1. health and fitness

Mental Health Treatment Study
Final Report
July 2011
Prepared for:
Social Security Administration
7111 Security Boulevard
Baltimore, MD 21244
Contract: SS00-05-60072
Prepared by:
Westat
1600 Research Boulevard
Rockville, Maryland
Investigators:
William D. Frey, Ph.D.
Robert E. Drake, M.D., Ph.D.
Gary R. Bond, Ph.D.
Alexander L. Miller, M.D.
Howard H. Goldman, M.D., Ph.D.
David S. Salkever, Ph.D.
Steven Holsenbeck, M.D.
Authors and Contributors
Listed below are the authors and contributors of the Final Report.
Authors
William D. Frey, Ph.D. (Co-Principal Investigator)
Westat
Mustafa Karakus, Ph.D.
Westat
Robert E. Drake, M.D., Ph.D. (Co-Principal Investigator)
Dartmouth Medical School
Roline Milfort, Ph.D.
Westat
Gary R. Bond, Ph.D.
Dartmouth Medical School
Jarnee Riley, M.S.
Westat
Alexander L. Miller, M.D.
University of Texas Health Science Center
at San Antonio
Cheryl Reidy, B.S.N., M.Ed.
Westat
Howard H. Goldman, M.D., Ph.D.
University of Maryland School of Medicine
David S. Salkever, Ph.D.
University of Maryland Baltimore County
Julie Bollmer, Ph.D.
Westat
Megan Collins, M.B.A.
Westat
Contributors
Susan T. Azrin, Ph.D.
National Institute of Mental Health (formerly at Westat)
David Judkins, M.A.
Westat
Deborah Becker, M.Ed., C.R.C.
Dartmouth Medical School
Hilary Kruger, M.A.
Westat
James Bethel, Ph.D.
Westat
Janice Machado, M.B.A.
Westat
Michael Cohen, M.A., C.A.G.S.
New Hampshire Alliance on Mental Illness
Troy Moore, Pharm.D., M.S.Pharm., B.C.P.P.
South Texas Veterans Health Care
System 119
David Garner
Westat
Brent Gibbons, M.P.P.
University of Maryland Baltimore County
Rene Gonin, Ph.D.
Westat
Linda Gowen, B.S.
Westat
Sigurd Hermansen, M.A.
Westat
Sheila O’Neill, M.S.W., L.C.S.W.
Thresholds
Jon Ratner, Ph.D.
Westat
Sandy Reese, B.A.
LifeWorks Northwest
Carina Tornow, M.A.
Westat
Acknowledgements
Successful completion of a study as large and complex as the Mental Health Treatment
Study (MHTS) requires extraordinary efforts from many individuals beyond those directly involved
in the day-to-day operations. The MHTS needed people who believed in it, supported it, and worked
behind the scenes to ensure its success. The study had many such individuals. The investigators
appreciate the efforts of these many people who contributed to the study’s success.
We are particularly appreciative of the support of the Social Security Administration,
especially Thomas Hale, Ph.D., the current Project Officer; Susan Kalasunas, the original Project
Officer; and Soosan Shahrokh, Ph.D., the Alternate Project Officer. Each one worked extremely
hard in many ways to support the study. Whether it required getting last minute approvals, following
through with individual beneficiary concerns with care, or asking probing questions to be sure that
the research team understood the nuances of the Social Security Disability Insurance (SSDI)
program, Tom, Susan, and Soosan were always ready to step in and provide guidance to facilitate
seamless study operations.
We also wish to thank the directors of the 23 study sites, their boards, and their staff whose
efforts and commitments to support the study and provide the best services possible to SSDI
beneficiaries—not their typical clientele—were invaluable. Research sites often do not fully recover
the actual costs of participating in studies such as the MHTS. For this reason alone, it takes an
extremely dedicated director to commit center staff and resources to participate in large research
studies of this nature knowing that there is an unrecoverable cost. The MHTS was no different. The
directors and staff of the community mental health centers (listed at the end of this report) deserve a
large share of the credit for helping to make the MHTS a successful study.
We greatly appreciate the efforts of ValueOptions/Colorado Health Networks and Hirsch
Financial Services for their support of the study through provision of health benefits to study
participants. This important feature of the intervention literally defined the term “access” to services
because on-time payments to providers are keys to continued services. Dedicated staff at these
organizations made certain that participant health care coverage met the standards set for the study.
We also thank the many computer systems, statistical, operations, and data management staff
at Westat who worked tirelessly to develop and maintain the study’s infrastructure which facilitated
the successful coordination of daily activities across 23 locations.
Finally, we are most grateful to the 2,238 SSDI beneficiaries who entrusted us with their care
by enrolling in the MHTS. No small amount of courage is required to commit to 24 months of
participation in a research effort—knowing that assignment to the treatment group is not a given.
Case-control studies similar to the MHTS usually experience large numbers of complaints and
exodus of participants assigned to the control group. It was just the opposite for the MHTS.
Participants in both the treatment and control groups often commented on how grateful they were
to be a part of the study, and how much they appreciated that the Social Security Administration was
studying ways to improve services to beneficiaries in the SSDI program.
Table of Contents
Chapter
Page
Executive Summary ..............................................................................................................
EX-1
Chapter 1: Study Overview
Background and Understanding .......................................................................
Design of the Mental Health Treatment Study ..............................................
Outcome Measures.............................................................................................
MHTS Investigators ...........................................................................................
1-5
1-12
1-14
1-15
Chapter 2: Study Design
Key Scientific Features.......................................................................................
Study Site Selection ............................................................................................
Sampling Methodology ......................................................................................
Sample Size ..........................................................................................................
Overview of the Enrollment and Randomization Process...........................
Intervention Benefits and Services...................................................................
Control Condition ..............................................................................................
Study Site Staffing ...............................................................................................
Fidelity Measures ................................................................................................
Outcome Measures.............................................................................................
Data Collection Procedures ..............................................................................
Interview Completion Rates .............................................................................
Data Preparation and Adjustments ..................................................................
Statistical Procedures..........................................................................................
Data Limitations .................................................................................................
2-1
2-3
2-6
2-9
2-10
2-12
2-15
2-16
2-17
2-18
2-21
2-23
2-24
2-28
2-30
Chapter 3: Enrollment and Participation
Are There Differences Between SSDI Beneficiaries Who
Enrolled and Those Who Did Not Enroll in the MHTS? ..............
Characteristics of the Beneficiaries Who Enrolled in the Study ..................
Why Did Eligible Beneficiaries Refuse to Enroll? .........................................
Discussion ............................................................................................................
3-2
3-10
3-17
3-18
Chapter 4: Outcomes
Employment Outcomes ....................................................................................
Earnings and Income .........................................................................................
Health and Quality of Life ................................................................................
Discussion ............................................................................................................
4-3
4-28
4-45
4-53
Table of Contents (continued)
Chapter
Page
Chapter 5: Implementation of Supported Employment and Other
Behavioral Health and Related Services
Background..........................................................................................................
Methods ...............................................................................................................
Results .................................................................................................................
MHTS Site-Level Fidelity-Outcome Analysis ................................................
Discussion ............................................................................................................
5-2
5-4
5-11
5-21
5-22
Chapter 6: Implementation of the Nurse Care Coordinator Role and
Systematic Medication Management
Background..........................................................................................................
Methods ...............................................................................................................
Results .................................................................................................................
Discussion ............................................................................................................
6-1
6-6
6-11
6-21
Chapter 7: Health Care and Supported Employment Financing
Health Care and SE Financing Principles .......................................................
Health Care Coverage and Financing Decisions............................................
Implementation of the HBP .............................................................................
SE Financing .......................................................................................................
HBP Expenditures..............................................................................................
Discussion ............................................................................................................
7-1
7-3
7-8
7-16
7-22
7-26
Chapter 8: Utilization of Services
Dependent Variables ..........................................................................................
Covariates.............................................................................................................
Estimation Models..............................................................................................
Results .................................................................................................................
Discussion ............................................................................................................
8-2
8-4
8-4
8-5
8-12
Chapter 9: Study Limitations, Key Findings, and Policy Implications
Study Limitations ................................................................................................
Key Findings .......................................................................................................
Policy Implications .............................................................................................
9-2
9-5
9-8
References ...................................................................................................................
Glossary of Terms ......................................................................................................
List of Acronyms ........................................................................................................
Acknowledgement of Study Site Staff ..................................................................
References-1
Glossary-1
Acronyms-1
Acknowledgement-1
Table of Contents (continued)
Table
List of Tables
Page
1-1
Number of SSDI beneficiaries under age 50 who had a primary
diagnosis of Other Mental Disorder: 1996-2009 ...........................................
1-7
2-1
MHTS sample by key stratifications ................................................................
2-8
2-2
Content domains for the Baseline, Quarterly, and final Followup
interviews .............................................................................................................
2-23
2-3
Interview completion rates by quarter and study group ...............................
2-24
3-1
Results of logistic regressions on MHTS participation (Enrolled = 1) ......
3-7
3-2
Mean beneficiary characteristics for quintiles based on predicted
enrollment probability of potential enrollees in the validation sample
(N=7,933) ............................................................................................................
3-11
3-3
Demographic and program characteristics of enrolled beneficiaries by
study condition....................................................................................................
3-13
3-4
Recent employment history and health status comparisons ........................
3-14
3-5
Physical health conditions identified in CAPI screener ................................
3-15
3-6
Total individual income .....................................................................................
3-17
3-7
Refusal frequencies .............................................................................................
3-17
4-1
Employment rates by treatment and control group ......................................
4-3
4-2
Employment rates by treatment and control group, stratified by age,
gender, diagnosis, and education ......................................................................
4-6
4-3
Other employment outcomes for all participants, participants who
worked at least one job, and participants who worked at least one
competitive job by treatment and control group ...........................................
4-7
Table of Contents (continued)
Table
4-4
List of Tables (continued)
Page
Statistically significant subgroup comparisons for other employment
outcomes including all participants, participants who worked at least
one job, and participants who worked at least one competitive job by
age, gender, diagnosis, and education ..............................................................
4-10
4-5
Classification of worker types by treatment and control group ..................
4-12
4-6
Characteristics of steady workers by treatment and control group .............
4-13
4-7
Types of occupations held by participants who reported working in a
job for each study-month by treatment and control group..........................
4-15
4-8
Types of occupations held by participants who reported working in a
job by diagnosis...................................................................................................
4-16
4-9
Types of occupations held by participants who reported working in a
job by education..................................................................................................
4-17
4-10
Potential predictor variables included in employment-related
regression models ...............................................................................................
4-18
4-11
Estimates of marginal effects on obtained employment, steady
worker, and number of months employed (for all participants)..................
4-20
4-12
Estimates of marginal effects on obtained employment, steady
worker, and number of months employed (for treatment group
participants only) ................................................................................................
4-22
4-13
Reasons for working part-time .........................................................................
4-23
4-14
Barriers to competitive employment for non-working treatment
group participants ...............................................................................................
4-25
4-15
Agreement with statements about Social Security Disability benefits
programs by treatment and control group ......................................................
4-27
4-16
Past month’s earnings (averaged over 8-post baseline interviews), past
three months’ earnings (at study exit) and earnings above SGA
(averaged over 8-post baseline interviews) by treatment and control
group.....................................................................................................................
4-29
Table of Contents (continued)
Table
List of Tables (continued)
Page
4-17
Number of quarters with past month’s earnings above, near, or
below SGA by treatment and control group ..................................................
4-31
4-18
Past month’s SSDI and SSI income (averaged over 8 post-baseline
interviews) by treatment and control group ...................................................
4-32
4-19
Past month’s income from other public programs (averaged over the
8 post-baseline interviews) by treatment and control group ........................
4-33
4-20
Past month’s total individual income (averaged over 8 post-baseline
interviews) and past month’s total household income (at study exit)
by treatment and control group ........................................................................
4-34
4-21
Past month’s earnings regression slope (averaged over 8 post-baseline
interviews) by treatment and control group ...................................................
4-35
4-22
Statistically significant earnings and income subgroup comparisons
for age, gender, diagnosis, and education for unconditional mean
comparisons, conditional mean comparisons, and percent reporting
non-zero amounts comparisons .......................................................................
4-36
Descriptive statistics for dependent measures included in multivariate
analyses of the MHTS impact on earnings and income by treatment
and control group ...............................................................................................
4-40
4-24
Estimates of average marginal MHTS effects on earnings...........................
4-44
4-25
Estimates of average marginal MHTS effects on income ............................
4-45
4-26
Mental and physical health status at baseline and study exit by
treatment and control group .............................................................................
4-46
4-27
Linear regression for mental health status at study exit (for all
participants) .........................................................................................................
4-47
4-28
Linear regression for mental health status at study exit (for treatment
group participants only) .....................................................................................
4-48
4-29
Satisfaction with life at baseline and study exit by treatment and
control group .......................................................................................................
4-48
4-23
Table of Contents (continued)
Table
List of Tables (continued)
Page
4-30
Ordered logistic regression estimates for quality of life (all
participants) .........................................................................................................
4-50
4-31
Ordered logistic regression estimates for quality of life (treatment
group participants only) .....................................................................................
4-50
4-32
Alcohol and drug use at baseline and study exit by treatment and
control group .......................................................................................................
4-52
4-33
Any drug use during the past 30 days at baseline and study exit by
treatment and control group .............................................................................
4-52
5-1
IPS service engagement rates for employed and not employed
treatment group participants in the last two reporting periods ...................
5-14
5-2
Percentage of treatment group participants who received IPS service
elements while enrolled in the MHTS .............................................................
5-14
5-3
IPS employment service contact rates of engaged but not employed
treatment group participants during the reporting period ............................
5-16
5-4
Rates of IPS contact with employed treatment group participants by
reporting period ..................................................................................................
5-17
5-5
Percentage of treatment group participants who received OBH or
related services while enrolled in the MHTS1 ................................................
5-18
5-6
Frequency and percentage of high and low access sites for other
behavioral health services ..................................................................................
5-19
5-7
Frequency and percentage of MHTS study sites that received various
overall quality of site-level IPS implementation ratings based on the
QMPD Site Reports: 2008 and 2009 ...............................................................
5-21
Frequency and percentage of MHTS study sites with various sitelevel barriers to implementation based on QMPD qualitative
assessments ..........................................................................................................
5-21
Site-level fidelity-outcome correlations in the MHTS ...................................
5-22
5-8
5-9
Table of Contents (continued)
Table
List of Tables (continued)
Page
6-1
Co-occurring psychiatric conditions for participants in the treatment
condition ..............................................................................................................
6-14
6-2
Frequency of co-occurring psychiatric conditions.........................................
6-14
6-3
Physical health conditions by source (treatment beneficiaries)
(N=981) ...............................................................................................................
6-16
6-4
NCC ratings of beneficiary engagement in SMM ..........................................
6-18
6-5
NCC ratings of prescriber engagement by prescriber location....................
6-18
6-6
Percent of beneficiaries seen by prescribers (at final study visit by
beneficiary) grouped by NCC ratings of prescriber engagement and
by prescriber location .........................................................................................
6-19
6-7
SMM implementation (% of participants) ......................................................
6-20
6-8
Summary of medication-related self-reports in final Followup
interview ...............................................................................................................
6-22
7-1
Insurance plans and estimated costs to Westat and participants .................
7-7
7-2
Number of participants enrolled in a new health plan by the WIP, by
year and by type of coverage .............................................................................
7-13
7-3
Medicare Part B premium changes ..................................................................
7-14
7-4
Distribution of HBP expenditures among all participants and 24month participants..............................................................................................
7-23
8-1
Weighted mean values and N's for treatment and control groups and
total: Overall utilization measures ....................................................................
8-6
8-2
Weighted mean values and N's for treatment and control groups and
total: ER and inpatient mental health utilization measures ..........................
8-7
8-3
Estimated average marginal treatment effects ................................................
8-8
Table of Contents (continued)
Table
List of Tables (continued)
Page
8-4
Estimated average marginal treatment effects for emergency room
visits and overnight hospital use for mental health problems .....................
8-11
8-5
Regression results for average main treatment effects with late
dropout dummy and dummy interaction with treatment* ...........................
8-12
Figure
List of Figures
Page
2-1
Map of MHTS study sites..................................................................................
2-5
3-1
Logic chart for assigning beneficiaries to enrollee group
classifications .......................................................................................................
3-4
4-1
Monthly employment rates for any job by treatment and control
group over the 24-month study period ...........................................................
4-4
4-2
Employment rates for any employment by treatment group and
control group and by study site ........................................................................
4-5
5-1
IPS fidelity for MHTS sites (based on 15-item IPS Fidelity Scale) .............
5-12
5-2
Site integration of IPS and behavioral treatment (IPS Fidelity Scale
item) ......................................................................................................................
5-13
6-1
Factors influencing prescriber medication decisions.....................................
6-4
6-2
Factors influencing prescriber medication decisions in SMM .....................
6-4
6-3
Role-based functions in SMM ..........................................................................
6-5
6-4
Concordance of SSA diagnostic group with SCID diagnosis (includes
dual diagnosis) .....................................................................................................
6-12
6-5
Concordance of SSA diagnostic group with SCID diagnosis (excludes
dual diagnosis) .....................................................................................................
6-13
6-6
Frequency of co-morbid physical health conditions for treatment
group beneficiaries..............................................................................................
6-15
Table of Contents (continued)
Figure
List of Figures (continued)
Page
6-7
Frequencies of beneficiaries with one or more physical health
conditions (N=981) ............................................................................................
6-17
6-8
BMI distribution in treatment group by gender.............................................
6-17
7-1
Distribution of total expenditures by type of claim including all
participants (n=1,121) (based on date of service) ..........................................
7-24
7-2
Distribution of cumulative expenditures by type of claim including all
24-month participants (n=982) (based on date of service) ..........................
7-25
7-3
Average HBP expenditures per person-per quarter (n=1,121)....................
7-25
7-4
Average HBP expenditures per participant month (n=1,121).....................
7-26
Appendix
List of Appendices
Page
2A
SF-12 Questionnaire and Scoring Algorithm .................................................
A-2A-1
2B
Imputation Procedures ......................................................................................
A-2B-1
3A
Variable Definitions for Enrollment and Participation Analyses ................
A-3A-1
3B
Descriptive Statistics for Participation Analysis Test and Validation
Samples.................................................................................................................
A-3B-1
3C
Probabilities of Enrollment for Test and Validation Samples .....................
A-3C-1
4A
Percent Obtained Employment (Any Job) by Site ........................................
A-4A-1
4B
Subgroup Comparisons for Employment Outcomes ...................................
A-4B-1
4C
Subgroup Comparisons for Earnings and Income Outcomes ....................
A-4C-1
4D
Regression Estimates for Formal Earnings ....................................................
A-4D-1
4E
Regression Estimates for Formal and Informal Earnings ............................
A-4E-1
4F
Regression Estimates for Income Variables ...................................................
A-4F-1
Table of Contents (continued)
List of Appendices (continued)
Appendix
Page
4G
Subgroup Comparisons for Health and Quality of Life Outcomes ............
A-4G-1
5A
IPS Fidelity Scale ................................................................................................
A-5A-1
6A
Listing of Medications and the Corresponding Physical Condition ...........
A-6A-1
6B
SCID Diagnostic Frequencies ..........................................................................
A-6B-1
6C
SMM QM Summary ...........................................................................................
A-6C-1
6D
NCC Survey Results ...........................................................................................
A-6D-1
6E
NCC Survey Administrator Version Results ..................................................
A-6E-1
6F
SMM Implementation Findings by Site...........................................................
A-6F-1
6G
SMM QA Findings by Site ................................................................................
A-6G-1
7A
Health Insurance Questionnaire.......................................................................
A-7A-1
7B
MHTS Monthly Encounter Form ....................................................................
A-7B-1
7C
Original and New Supported Employment and Related Services
Payment Schedules .............................................................................................
A-7C-1
Supplemental Appendix
List of Supplemental Appendices
Page
A
Study Implementation ........................................................................................
SA-A1
B
CAPI Screener.....................................................................................................
SA-B1
C
Baseline Questionnaire ......................................................................................
SA-C1
D
Treatment Quarterly Questionnaire ...............................................................
SA-D1
E
Control Quarterly Questionnaire ....................................................................
SA-E1
F
Final Followup Questionnaire .........................................................................
SA-F1
Note to Reader
This report aims to provide enough information to the reader to understand the purpose, design,
general procedures, outcomes, and conclusions of the Mental Health Treatment Study.
To aid understanding, the reader will find lists of acronyms and terminology used throughout the
chapters at the back of this report.
It was not possible to include all of the details of the study in one volume. Therefore, a number of
documents appear in the Appendices or in the Supplemental Appendices. The Appendices are a
valuable complement to this Final Report, and readers may wish to refer to them when reading the
report. The Appendices comprise a separate document and include expanded analyses, frequencies,
distributions, scales, forms, etc. referenced in the chapters. While the Supplemental Appendices
(also a separate document) provide interesting and detailed information about study implementation,
they are not required to comprehend this Final Report. The Supplemental Appendices include
details of the study implementation process and procedures, and copies of the study questionnaires.
Executive Summary
The Mental Health Treatment Study (MHTS) provides a test of the hypothesis that access to
supported employment (SE) services and systematic medication management (SMM) services,
coupled with the removal of some known programmatic disincentives, will enable Social Security
Disability Insurance (SSDI) beneficiaries with schizophrenia or an affective disorder to return to
work. Fielded by the Social Security Administration (SSA) between November 2006 and July 2010,
the test included 2,238 SSDI beneficiaries in 23 study sites throughout the United States.
Beneficiaries volunteering to participate in the study received a random assignment to either a
treatment group or a control group following completion of a comprehensive Baseline interview.
Beneficiaries participated for 24 months. The study collected self-report data quarterly on the
primary outcome measures of employment (including earnings), health status, and quality of life.
These data provided the means to test the hypothesis that access to evidence-based behavioral
health treatments and employment supports would result in improved outcomes.
Background
SSDI is a long-term disability insurance program designed to provide income to people who are no
longer able to work because of a disability. Age at the time of onset of disability determines the
number of quarters a person must have previously worked in order to qualify for the program. The
SSDI program addresses the income support and health care needs of workers who can no longer
engage in substantial gainful activity (SGA) due to a medically determinable physical or mental
impairment. Beneficiaries accepted into the program receive cash benefits based on their past job
earnings and, after a 2-year waiting period, are eligible for Medicare.
A high priority at SSA is to support SSDI beneficiaries in their efforts to return to work. In doing
so, the agency provides a broad range of programs and supports to facilitate a beneficiary’s pursuit
of employment goals. For example, the Ticket to Work program assists beneficiaries to gain access
to services needed to get and keep a job. The Trial Work Period (TWP) allows beneficiaries to test
their ability to work for up to 9 months over a rolling five-year period without worry of losing cash
benefits. The Expedited Reinstatement rule serves as a safety net for those beneficiaries who return
to work, get off the program completely, but then find at a later date that they can no longer work.
The Extended Period of Eligibility (EPE) begins immediately following the TWP and extends for 36
months. The program allows the beneficiary to resume receiving benefits if earnings fall below SGA
Executive Summary
EX-1
in a given month. Programs and supports like these assist SSDI beneficiaries to achieve their
employment goals. The MHTS is just one in a series of demonstrations at SSA to explore new and
improved employment supports to beneficiaries in the SSDI program. Other efforts include the
Benefit Offset National Demonstration (BOND) project and the Youth Transition Demonstration
(YTD).
The MHTS developed out of three pressing realities. First, beneficiaries with mental disorders in the
SSDI program, specifically those with psychiatric impairments, have been and continue to be a
significant policy concern. The percentage of beneficiaries under the age of 50 who receive SSDI
benefits due to a psychiatric impairment has steadily increased over the past 10 years and continues
to grow. Second, evidence-based employment supports exist with demonstrated success in helping
people with mental illness return to work. There is also a large body of evidence demonstrating that
antipsychotic medications, mood stabilizers, and antidepressants provide symptom relief for persons
with schizophrenia, bipolar disorder, or depression. There is virtually universal use of medication to
treat these conditions in public and private mental health facilities in the United States. Third,
surveys of individuals with severe mental illness consistently indicate that they want to work. Yet
they have one of the lowest employment rates of any subpopulation.
Taken together, the concerns about supporting beneficiaries in the SSDI program, the promise of
evidence-based mental health treatments and employment supports, and lack of understanding how
these treatments and supports might work with SSDI beneficiaries with psychiatric impairments,
prompted SSA to test the effectiveness of the treatment intervention. The policy concerns discussed
above drove the following research questions, which the study was expected to answer:
1.
To what extent does delivering appropriate mental health treatment and employment
supports lead to better employment, health status, and quality of life among SSDI
beneficiaries with schizophrenia or an affective disorder?
2.
What programmatic disincentives exist that create barriers for Title II beneficiaries with
schizophrenia or an affective disorder to return to work?
3.
What specific programmatic changes would support the efforts of SSDI beneficiaries
with schizophrenia or an affective disorder to sustain competitive employment?
Executive Summary
EX-2
Study Design
The MHTS was a randomized controlled trial of SSDI beneficiaries with schizophrenia or an
affective disorder. This feature involved randomly assigning beneficiaries enrolling into the study to
either a treatment group or a control group. The treatment group received a comprehensive package
of services and benefits, including evidence-based SE, SMM, behavioral health and related services,
and comprehensive insurance to pay for needed services and out-of-pocket expenses. In addition,
SSA suspended each beneficiary’s medical Continuing Disability Review (CDR) for a period of 3
years from the date of study enrollment. The control group received a resource manual that listed
available local and national services and resources for persons with mental illness, and a nominal
payment of $100 for participating in quarterly interviews. The control group was still subject to the
medical CDR.
The MHTS followed the intent-to-treat (ITT) model so that outcomes for all study participants
would be included in the analysis, regardless of the extent to which they participated in the treatment
or completed study activities. The ITT approach is most appropriate for answering the study
questions within the context of a policy concern about expected outcomes of individuals offered
access to services, and not simply expected outcomes of individuals who use services at a particular
level of engagement and intensity. The ITT analysis is conservative and works against finding
positive effects, but it is the most appropriate design for evaluating the impact of a program that
would offer—but not require—engagement in services.
Twenty-three study sites located in each region of the U.S., except the Southwest, participated in the
study. Two criteria dictated site selection, including: (1) the ability to deliver behavioral health and
SE services and (2) documented fidelity to the Individual Placement and Support (IPS) model for
SE. All but two study sites were community mental health agencies that provided an array of mental
health services. The remaining two sites included a vocational center and a housing center for
homeless people with mental illness that also provided vocational services. The vocational center
contracted mental health services with the county mental health facility. The homeless program
brokered mental health services in the nearby community.
SSDI beneficiaries eligible to participate in the MHTS included those with a primary diagnosis of
schizophrenia or an affective disorder, between the ages of 18 and 55 (inclusive), and residing within
a 30-mile radius of one of the study sites. Beneficiaries excluded from consideration for the study
were those living in a nursing home or other custodial institution, had a legal guardian, had a lifethreatening or terminal physical health condition, were already receiving SE services from their
Executive Summary
EX-3
participating demonstration site within 30 days of recruitment, or were currently working in a
competitive job within 6 months of recruitment.
Data sources for the MHTS included information collected during the study and data retrieved from
external sources. The primary data source used to test the stated hypothesis were self-report
measures collected directly from enrolled study participants in the treatment and control groups
through computer-assisted interviews. Research staff interviewed each treatment and control group
beneficiary at enrollment (i.e., baseline) and then every 3 months following baseline, resulting in
seven Quarterly interviews and a final Followup interview over the 24-month period of
participation. The nine interviews collected information about the beneficiary’s demographic
characteristics, work history and employment, income, health status, alcohol and drug use, health
care coverage, health care service utilization, understanding of SSA’s return-to-work programs, and
quality of life. Additional information collected on the treatment group only included data related to
pre-treatment activities, receipt of SE services, SMM services, and other behavioral health services,
as well as expenditures related to service utilization.
Staff at each study site included a dedicated Nurse Care Coordinator (NCC) and Research Assistant
(RA), funded by the study. The NCC served in dual roles supporting both treatment (particularly the
SMM component) and research (completing various reports and data-entry forms). The RA served
primarily in a research role and was responsible for recruitment, enrollment, completing the
interviews, and other research tasks. In addition to the NCC and RA, each site required an adequate
number of SE specialists to serve the treatment group participants.
Standard procedures guided study recruitment activities, including introductory mailings, recruitment
telephone calls, and Recruitment Information Group (RIG) meetings. Once beneficiaries attended
two required RIG meetings, they enrolled in the study by completing a screener (to assess
competence to give consent, confirm eligibility, and then provide consent), and completing the
Baseline interview. A computer-generated call-in program randomized each beneficiary to either the
treatment or control group following completion of the Baseline interview.
For beneficiaries randomized to the treatment group, pre-treatment activities included completion of
the Axis 1 Disorders section of the Structured Clinical Interview for the Diagnostic and Statistical
Manual of Mental Disorders, fourth edition (referred to as the SCID), the Brief Assessment of
Cognition in Schizophrenia, and a General Medical Exam (GME). Additionally, beneficiaries
randomized to the treatment group were assigned treatment providers (e.g., prescriber, SE specialist,
therapist, or case manager) as needed.
Executive Summary
EX-4
Throughout the study, treatment group participants had access to evidence-based mental health
treatments and services, as well as integrated mental health services and employment support
services. The treatments and services offered to participants included SE, SMM, and other
behavioral health (OBH) and related services; regular meetings with the NCC; benefits counseling
(where possible); and payments for out-of-pocket expenses for behavioral health care (health
insurance coverage, debit card for payment of prescriptions, co-pay reimbursement, etc.).
All treatment group participants who remained in the study at 20 months participated in the
development of a personal transition plan. The 24th month was the end point for completing the
transition plan. The primary goal of the transition plan was to maximize the potential for the
participant to maintain all positive employment, behavioral health, and other outcomes by ensuring
continued access to care and services after exiting the study.
The study design included a range of quality assurance procedures built into the various components
of the study to ensure that the treatment group received high quality SE, OBH, and related services.
Specifically, three Quality Management Program Directors (QMPDs) worked closely with staff at
each of the 23 study sites to ensure participants received evidence-based treatments. In addition,
Nurse Care Coordinators (NCCs) completed the SE/OBH and SMM quality management reports
which collected data on receipt of services. The QMPDs and SMM experts often consulted directly
with site staff to address issues as they arose. Other quality assurance activities ensured appropriate
execution and use of study benefits (e.g., examinations of MHTS debit card use and the provision of
procedural manuals and periodic staff trainings throughout the course of the study).
To ensure protection of study participants, the MHTS included a number of safeguards. First,
voluntary withdrawal meant termination of all study benefits for treatment group participants. A
standard withdrawal protocol assisted this process and ensured that no harm came to participants
electing to withdraw from the study. A standard “administrative drop” process guided staff in
removing treatment group participants from the study who did not obtain the required GME or
failed it. Finally, event-generated reports of any adverse event or unanticipated problem were
submitted to the MHTS Institutional Review Board (IRB) Administrator and SSA as needed.
Assessment of the Randomization Procedure. A critical aspect of the participation analyses was
an assessment of the success of the randomization procedure in creating two equal groups in the
treatment and control arms of the study. The results of tests revealed that the procedure successfully
created two equal groups on the key measures at baseline. These tests found no differences between
Executive Summary
EX-5
the treatment and control groups on the (1) amount of work in the past two years, (2) mental health
as measured by the SF-12, (3) physical health as measured by the SF-12, or (4) total individual
income in the past month.
Findings
The following sections address the study findings. Section A presents findings about Enrollment and
Participation, including an estimate of how many SSDI beneficiaries might take advantage of an
MHTS-like program, and the reasons beneficiaries refused to enroll in the study when offered the
opportunity. Section B addresses the Primary Study Outcomes, including employment rate, health
status, and quality of life. Section C presents findings associated with the Secondary EmploymentRelated Outcomes. Section D presents findings associated with the Secondary Earnings and Income
Outcomes. Section E addresses study implementation of SE and SMM. Section F presents findings
from the Health Benefits Plan (HBP). Finally, section G presents findings related to Utilization of
Services.
A. Enrollment and Participation
1.
Based on the sample of potential enrollees, the findings suggest that nearly 14 percent of
SSDI beneficiaries with schizophrenia or an affective disorder (who can be reached by
telephone or letter) would take advantage of an MHTS-like program.
2.
Some of the strongest predictors of enrollment were those related to prior work activity
reported in the SSA administrative records, including beneficiaries who reported
earnings within the past six months (prior to enrollment), had assigned their Ticket (in
the Ticket to Work program), or had a Trial Work Period End Date within the past
three years.
3.
Beneficiaries enrolling in the study tended to be younger, on the SSDI rolls a shorter
period, only on SSDI (as opposed to concurrent SSI beneficiaries), and did not have a
Representative Payee.
4.
The most commonly recorded single reason for beneficiaries not enrolling in the study
was general disinterest (37%). However, nearly 40 percent of all beneficiaries for whom
a reason was recorded, indicated work-related or physical health issues as the reason
they did not enroll in the study.
Executive Summary
EX-6
B. Primary Study Outcomes: Employment rate, physical health status, mental health
status, and quality of life
5.
The 24-month employment rate for the treatment group was 61 percent compared to
40 percent for the control group. The difference between these percentages was
significant (p-value < 0.001).
6.
Measured at baseline and again at study exit, the treatment group showed a significant
improvement over the control group in mental health status (p-value < 0.001).
7.
Measured at baseline and again at study exit, both the treatment and control groups
showed a slight decline in physical health status. The changes between the groups were
not significant (p-value = 0.924).
8.
Measured at baseline and again at study exit, the treatment group showed a significant
improvement over the control group in quality of life (p-value < 0.001).
C. Secondary Outcomes: Employment-related
9.
Treatment group participants had significantly better outcomes than did control group
participants on total months employed (d = 2.58, p-value < 0.001), consecutive months
of employment at study exit (d = 1.43, p-value < 0.001), average weekly earnings at
main job (d = $40.54, p-value < 0.001), average hours per week at main job (d = 4.29, pvalue < 0.001), and highest hourly wage (d = $2.56, p-value < 0.001).
10.
When the treatment and control participant groups were narrowed to participants who
worked at least one job, the findings revealed significant differences between the
treatment and control groups on total months employed (d = 0.69, p-value = 0.016),
consecutive months of employment at study exit (d = 0.87, p-value = 0.001), and
average weekly earnings at main job (d = $4.07, p-value = 0.080). Neither average hours
per week at main job (d = 0.76, p-value = 0.173) nor highest hourly wage (d = $0.02, pvalue < 0.443) were significant.
11.
When the participant groups were narrowed to participants who worked at least one
competitive job, the findings were not significant, with one exception. Only total
months employed (d = 0.91, p-value = 0.017) was significant, favoring the treatment
group.
12.
The investigators classified study participants into three groups: steady workers (those
who worked 10 or more months), erratic workers (those who worked between 3 and 9
months), and minimal workers (those who worked less than 3 months including those
who did not work at all). A Chi-square test comparing differences in classifications
between the treatment and control groups found that the treatment group had a
significantly greater number of study participants in the steady worker and erratic
worker classifications (and consequently lower numbers in the minimal worker
classification) than did the control group (p-value < 0.001).
13.
Logistic regression predictions of obtained employment and steady worker status
identified five common predictors, including the treatment dummy (ME = 0.226, p-
Executive Summary
EX-7
value < 0.001; ME = 0.139, p-value < 0.001 respectively). Additional predictors
included baseline physical health status (p-values < 0.001, = 0.021 respectively), worked
in the last 2 years (p-values < 0.001, < 0.001 respectively), unemployment rate (p-values
= 0.017, = 0.002 respectively), and months on rolls (p-values = 0.005, < 0.001
respectively). A zero inflated negative binomial regression on number of months
employed identified the same five significant predictors, including the treatment dummy
(ME = 2.57, p-value < 0.001), baseline physical health status (p-value < 0.001), worked
in the last 2 years (p-value < 0.001), unemployment rate (p-value < 0.001), and months
on rolls (p-value < 0.001). This model identified three additional predictors, including
ever had a ticket (p-value = 0.031), baseline mental health status (p-value = 0.020), and
number of hospital stays at baseline (p-value = 0.016).
14.
Logistic regression predictions of obtained employment and steady worker status for
the treatment group only identified the same predictors of employment-related
outcomes. However, the models revealed that SE engagement (p-value = 0.002, p-value
< 0.001 respectively) and receipt of case management services (p-value = 0.004, p-value
= 0.004 respectively) were also significant. The zero inflated negative binomial
regression prediction of number of months employed also identified SE engagement
(p-value < 0.001) and receipt of case management services (p-value = 0.031) as
significant predictors of number of months employed. This model identified three
additional predictors, including ever had a ticket (p-value = 0.031), baseline mental
health status (p-value = 0.020), and number of hospital stays at baseline (p-value =
0.016).
15.
Among the five general industry job type categories, service occupations, and sales and
office occupations accounted for more than 70 percent of all jobs obtained by study
participants. This figure remained steady at baseline, during the study, and at the end of
the study. Looking at the overall occupational data, treatment group and control group
participants entered service occupations (37%, 37% respectively), and sales and office
occupations (36%, 34% respectively) in roughly equal percentages. Neither difference
between the treatment and control groups was statistically significant. In roughly equal
percentages, both the treatment and control groups entered into management, business,
science and arts occupations (13%, 14% respectively), natural resources, construction,
and maintenance occupations (3%, 2% respectively), and production, transportation,
and material moving occupations (11%, 13% respectively).
D. Secondary Outcomes: Earnings and income
16.
Analyses of average earnings (past month earnings averaged over eight post-baseline
interviews) show that treatment group participants earned significantly more than did
the control group participants. The first analysis concerns comparisons between the
unconditional means (including all 2,238 study participants) of the treatment and
control groups ($148 vs. $97, p-value < 0.001). A second analysis concerns the
comparison between conditional means (i.e., including only those participants with nonzero earnings) of the treatment and control groups ($251 vs. $228, p-value < 0.001). The
third analysis concerns the percentages of participants in the treatment and control
groups with earnings (59% vs. 43%, p-value < 0.001).
Executive Summary
EX-8
17.
Analyses of past three months earnings (from the final Followup interview only) show
that treatment group participants earned significantly more than did the control group
in two of three analyses. The first analysis concerns comparisons between the
unconditional means (all study participants) of the treatment and control groups ($859
vs. $479, p-value < 0.001). The second analysis concerns the comparison between
conditional means (i.e., including only those participants with non-zero earnings) of the
treatment and control groups ($2,538 vs. $2,739, p-value = 0.607). The third analysis
concerns the percentages of participants in the treatment and control groups with
earnings (34% vs. 17%, p-value < 0.001).
18.
Eight percent of the study participants showed average earnings over the 24-month
study period that exceeded the current level of SGA. Beneficiaries in the treatment
group did not experience an increase in work that SSA considers SGA when compared
to participants in the control group. Neither did participants in the treatment group
experience a reduction in benefit payments when compared to participants in the
control group.
19.
Analyses of past month’s individual income (averaged over eight post-baseline
interviews) and past month’s household income (at study exit only) show significant
treatment effects. The first analysis shows unconditional means (d = $59.56, p-value <
0.001; d = $16.74, p-value = 0.063 respectively); the second analysis shows conditional
means (d = $59.56, p-value < 0.001; d = $23.39, p-value = 0.043 respectively.
E. Implementation of SE and SMM
20.
Eighty percent or more of the study sites achieved a high level of IPS program
implementation (i.e., met the documented standard for high fidelity). This high level of
implementation persisted across the entire study period.
21.
The level of unengagement in employment services among treatment group participants
was relatively low overall (~10 %).
22.
Concordance between the SSA diagnostic category and the SCID diagnosis was greater
than 80 percent.
23.
More than 87 percent of treatment group participants had at least one physical health
condition, and 69 percent had two or more. More than half of all beneficiaries in the
treatment group had a Body Mass Index (BMI) in the obese range.
24.
The quality of SMM varied considerably across study sites. The off-site location of
many prescribers affected this measure. For purposes of conducting the MHTS, the
decision to allow participants to remain with their off-site prescribers was reasonable.
However, the goal of MHTS was to deliver an integrated package of services to
participants. Having off-site prescribers presented great difficulties in integrating the
SMM components with other treatments.
Executive Summary
EX-9
F. Health Benefits Plan
25.
Spending through the Health Benefits Plan averaged $6,986 per study participant per
year. Spending was less ($6,342 per year) for the overall treatment group population
(which includes those participating for less than 24 months). These figures do not
include spending for services related to the NCC role.
26.
The overall spending distribution included over 70 percent for SE services, 11 percent
for health insurance premiums, 8 percent for medication prescriptions, 7 percent for
behavioral health services, and less than 3 percent for employment-related work
expenses, transportation, and other miscellaneous expenses. SE services were the only
services completely paid by the study.
G. Utilization of Services
27.
The treatment intervention had significant positive impacts in reducing inpatient
hospital use (for both admissions and number of days) and psychiatric crisis visits.
28.
As expected with the intervention package, treatment group participants showed
significantly higher use of regularly scheduled clinic or mental health visits.
29.
The average reduction in hospital days was 0.9 days per year which translated into
approximately $1,800 per year per person. Sustaining these costs over longer periods
would increase their magnitude.
SSDI beneficiaries with schizophrenia or an affective disorder who indicated that they wanted to
return to work and who enrolled in the study and then received a random assignment to the
treatment intervention, as a group, ended the study with significantly better employment rates, better
mental health, and a higher quality of life. Further, they ended the study with many additional
significantly better results than the control group, including higher earnings and income, more hours
worked, a greater number of months worked, and greater satisfaction with their main job. The
treatment intervention essentially delivered to SSA everything it asked for in the study design. For
whatever reason(s), the intervention package, which comprised a rich mix of services and benefits,
was successful in getting a large portion of the beneficiaries who enrolled in the study into jobs—the
primary goal of participation. The next section explores the implications of these findings.
Policy Implications
The original research question concerns the extent to which the treatment intervention resulted in
better outcomes. The findings presented above easily answer this question. The remaining two
questions about programmatic barriers and specific programmatic changes do not come easily or
directly from the study data. Instead, they are qualitative in their nature. The following questions and
Executive Summary
EX-10
answers attempt to provide insight that will provide the agency with the information it needs to
move the MHTS result forward.
1.
Is there significant interest among SSDI beneficiaries with schizophrenia or an
affective disorder to access an MHTS-like program?
Answer. Yes. Today there are nearly 2.25 million SSDI beneficiaries with a psychiatric impairment.
The largest portion of this group includes those with schizophrenia or an affective disorder. The
study findings suggest that a reasonable take-up rate would be nearly 14 percent or 306,000 of the
2.25 million SSDI beneficiaries with a psychiatric impairment. In addition, if SSA targeted
enrollment to only those with recent work activity (e.g., recent TTW activation), the rate would
range between 26 and 32 percent.
2.
What essential services or features of services are required to achieve MHTS-like
results?
Answer. The following services and features comprise the unique characteristics of the MHTS:
3.
a.
Essential services: Evidence-based SE (the IPS model), SMM, OBH services (such as
therapy, counseling, substance abuse counseling, etc.), benefits counseling, and modest
monetary support to pay for services that are otherwise inaccessible.
b.
Community mental health centers (CMHC): CMHCs offer both the comprehensive
range of services needed to treat mental illness and provide integrated vocational
services in the community mental health center. The experience of the MHTS suggests
that the community mental health center was ideal for delivery of these essential
services.
c.
Care coordinator: Use of a care coordinator to facilitate SMM was a key feature of the
mental health services provided in the MHTS. The SMM program in the MHTS used
the NCC to facilitate and promote prescriber use of evidence-based guidelines and
recommendations for medication management of severe and persistent mental illnesses.
d.
Out-of-pocket expenses: Payment for out-of-pocket mental health and essential workrelated expenses was an important feature that improved access to services and jobs for
study participants.
How much did the study spend on each treatment group participant to achieve the
study results?
Answer. Overall, the MHTS spent an average of $6,986 per year per participant. These expenditures
do not reflect all of the participant costs for services needed during participation in the study. NCC
services, for example, were paid through the study and not as part of the HBP. Medicare or
Medicaid paid for many of the mental health and general health services required by participants.
Executive Summary
EX-11
Expenditures through the HBP made it clear that obtaining payment for all behavioral health care
costs, along with payment of insurance premiums, removed some of the putative barriers to
participation in active efforts to return to work for individuals in the treatment group. These
payments represented less than 10 percent of the total HBP expenditures, suggesting that across all
participants in the treatment group, gaining access to needed health care amounted to approximately
$53 per month. This is a remarkably low cost for improving access to needed health care.
It is noteworthy that fewer hospital stays and fewer psychiatric emergency or crisis visits may offset
some of the HBP expenditures. Further exploration of the MHTS data may clarify the potential for
health care.
4.
What programmatic disincentives exist that create barriers for Title II beneficiaries
with schizophrenia or an affective disorder to return to work?
Answer. Three general barriers describe the kinds of problems SSDI beneficiaries faced in gaining
access to needed health care and employment programs and supports.
Under current conditions, SSDI beneficiaries have insufficient access to health
care programs, services, and treatments. Each of the problems presented below
create distinct barriers to effective treatment, better functioning and, ultimately, to
employment. No one barrier by itself is particularly troublesome. However, as a group,
they represent formidable obstacles to return to work.
a.
i.
The study revealed many instances where insufficient access to programs, services,
and treatments was a problem. Approximately 7 percent (74) of treatment group
participants required enrollment in one or more of Parts A, B, or D of Medicare.
ii.
Reports from participants in the treatment group suggested that in the past, the cost
of insurance co-pays (for health care visits, prescription medications, etc.) kept some
participants from seeking treatment. In general, it is clear that such decisions to seek
treatment were complicated and highly individual. For example, a participant related
a story that prior to the study she chose not to refill a prescription for her psychiatric
medication because it was too expensive given her financial condition. However,
upon further discussion, it became clear that the decision was more complicated than
that. She felt the medication was not effective (had unpleasant side effects, was not
sufficiently reducing symptoms) and, therefore, was not worth the cost of the co-pay
to refill it. Many previous trips to the doctor to find a better medication proved
expensive and ineffective. It was no longer worth the effort required to seek a
solution.
iii.
One ongoing concern throughout the study was the role that high-cost psychiatric
medications played in finding effective treatments for participants in the treatment
group. Several situations were problematic. One clear problem was the effect of the
Executive Summary
EX-12
so-called “donut hole” on medication use. NCCs reported that prior to the MHTS
when participants came up against the “donut hole,” they would stop refilling their
prescription(s) for psychiatric (and other) medication(s) because the high costs did
not fit within their budget. Another problem reported by some participants was that
prior to the study, prescribers were reluctant to prescribe some (potentially more
effective) medications due to the high cost to the patient.
iv.
Another problem was finding a Part D plan that covered the preferred medications.
During the study, this always presented a challenge, and required more efforts from
the Westat Insurance Planner than any other insurance issue. A change in the Part D
prescription medication plan required waiting until the open enrollment period. The
findings would not likely be the same without the support of the study to pay for
medications when they were not covered or to pay for high co-pays.
v.
The study paid special attention to receipt of mental health case management
services because case management is a key for achieving adequate mental health
treatment. The fact that only 54 percent of participants received mental health case
management, and of these, 28 percent received their case management services offsite, is far below the expected rate in high fidelity IPS programs serving clients with
severe mental illness.
b.
Under current conditions, SSDI beneficiaries lack access to evidence-based SE
services in community mental health centers. Analysis of the health services
utilization data reveals that the treatment group used significantly more vocational
services than the control group. This was expected, given the intervention package
included the provision of evidence-based IPS SE. These services, which emphasized
competitive employment, clearly contributed to the positive employment results
attained in the study. However, as discussed elsewhere in this report, community mental
health centers do not generally provide SE services. Payment for these services is
limited to Medicaid. In fact, even Medicaid does not cover these services in every state.
c.
Many SSDI beneficiaries with schizophrenia or an affective disorder have
complex co-morbid physical conditions that impede efforts to return to work.
Physical impairments created by a wide variety of health conditions were serious
deterrents to employment efforts made by study participants, NCCs, and employment
specialists. As reported above, 87 percent of the participants in the treatment group had
at least one co-morbid physical health condition, 69 percent had two or more, and more
than half were obese. Frequent comments on conference calls throughout the study
noted that the SSDI beneficiary population in the study was much less healthy and had
more problematic health conditions than did the populations normally served by the
mental health centers. These comments corroborated the data collected from treatment
group participants and formal documentation of why some beneficiaries do not return
to work. In fact, good physical health (as measured by the SF-12) frequently came up as
a predictor of employment (obtained employment, steady worker, or number of months
employed), suggesting that there was enough variability in physical health to pick it up
in multivariate analyses.
Executive Summary
EX-13
Chapter 1
Study Overview
The Mental Health Treatment Study (MHTS) provides a test of the hypothesis that access to
supported employment (SE) services and systematic medication management (SMM) services,
coupled with the removal of programmatic disincentives, will enable Social Security Disability
Insurance (SSDI) beneficiaries with schizophrenia or an affective disorder to return to work.
Following nearly 6 years of conceptual development, panel reviews and preparation, the Social
Security Administration (SSA) fielded the study between November 2006 and July 2010.
The study design was a randomized controlled trial that included 2,238 SSDI beneficiaries between
the ages of 18 and 55 with a primary impairment of schizophrenia or an affective disorder.
Following their election to enroll in the study (giving consent and then completing an extensive
Baseline interview), study participants received a random assignment to either a treatment or control
group. Participants in both groups subsequently completed seven Quarterly interviews and a final
Followup interview, designed to collect self-reported data on the outcome of interest.
The treatment group received a comprehensive package of benefits that included SE services, SMM,
behavioral health and related services (and comprehensive insurance to pay for them),
reimbursement for out-of-pocket behavioral health expenses, and a guarantee that the SSA would
suspend their medical Continuing Disability Reviews (CDRs) for 3 years from the date of study
enrollment. The study paid all costs associated with obtaining services and prescription medications
associated with behavioral health care that were not paid for by other sources.
The control group received “services as usual” with minimal enhancements, which included a
comprehensive manual of services that were available in their community and a modest payment
amounting to $100 over the study period for completing the interviews. Medical CDRs remained in
effect for these beneficiaries. Beneficiaries in the control group were not limited or restricted in any
way from seeking community services to achieve their employment goals. In fact, they were
encouraged to use the manual provided to them to seek assistance in obtaining employment.
Services as usual refer to the fact that beneficiaries in the control group receive no special or unique
services from the study. Any services they seek or receive depend entirely upon their own efforts.
Chapter 1: Study Overview
1-1
Beginning on the day of randomization, enrollees remained in the study for a period of 24 months.
At the 20th month of study participation, beneficiaries in the treatment group participated in a
formal transition process designed to ease them back into “services as usual,” without the
enhancements provided by the intervention.
A core component of the treatment intervention was the evidence-based Individual Placement and
Support (IPS) model of SE services. The seven core principles that form the foundation of the IPS
model include the following:
1.
Consumer choice. Individuals who are interested in work are eligible for IPS without
exception.
2.
Integrated services. Vocational and mental health services together are a part of the overall
treatment approach.
3.
Competitive employment in regular work settings. No pre-employment training or placement in
sheltered or segregated work settings is required before placement in competitive work
settings.
4.
Rapid job search. IPS employment specialists help clients begin the job search soon after
enrollment in IPS.
5.
Personalized follow-on support. After placement, individuals and their employers (if desired)
receive ongoing support for as long as they need it.
6.
Person-centered services. Clients’ personal preferences, experiences, strengths, and choices
drive the job search and follow-on supports rather than the judgment of the
employment specialist.
7.
Benefits counseling. Clients receive benefits counseling to ensure successful navigation of
any impact of employment on government entitlements such as Medicaid or Social
Security benefits.
Notable among this list are the principles of consumer choice and rapid job search. Contrary to many
employment programs for persons with mental illness, IPS places no limitations or restrictions on
who is eligible to participate. For example, there is no programmatic requirement to meet a
particular therapeutic goal prior to work placement, as is the case in many mental health treatment
programs. Any individual interested in work is eligible for the program. With the principle of rapid
job search, individuals enter the labor force as soon as they feel ready. There is no special preparation
through training or career exploration, as is the case with some vocational programs. When the
individual feels ready, they begin work. Any program adopting the evidence-based IPS model must
Chapter 1: Study Overview
1-2
demonstrate fidelity to the seven core principles if it expects to achieve the measured outcomes
(Bond, 2004; Loveland, Driscoll, & Boyle, 2007).
Another core component of the treatment intervention was the evidence-based SMM services.
While psychiatric symptom relief by itself typically may not lead to return to employment for those
with a severe mental illness, it is clear that symptoms of depression, mania, and psychosis can
interfere with functionality and interpersonal relations. Hence, symptom amelioration and stability
may contribute to the success of behavioral interventions, such as SE, and to greater productivity
and satisfaction with life (Awad et al., 1999; Leidy et al., 1998; Skevington et al., 2001). While both
IPS SE services and SMM are widely recognized as having a strong evidence base, investigators have
never fully tested these treatment modalities as a combined intervention package. Additionally, prior
studies of the effectiveness of IPS SE or SMM include samples of people with severe mental illness
who were already engaged in treatment when approached to participate in the research. By contrast,
people recruited to participate in the MHTS came from the rolls of SSDI beneficiaries and not all of
them were currently engaged in treatment. Hence, SSA wants to know if it is effective to offer this
package of services in community mental health centers across the United States, and whether
providing SSDI beneficiaries with access to these services improves employment, health, and quality
of life.
SSA’s Request for Proposals (SSA-RFP-05-1044) posed three research questions for the MHTS.
They were as follows:
1.
To what extent does delivering appropriate mental health treatment and employment supports
lead to better employment, health status, and quality of life among SSDI beneficiaries with
schizophrenia or an affective disorder?
2.
What programmatic disincentives exist that create barriers for Title II (SSDI) beneficiaries
with schizophrenia or an affective disorder to return to work?
3.
What specific programmatic changes would support the efforts of SSDI beneficiaries with
schizophrenia or an affective disorder to sustain competitive employment? (SSA, 2005, p. 6)
In consultation with SSA, MHTS investigators interpreted the research questions in more
operational terms. The first question concerns the primary outcomes of the study, including the
extent to which the intervention results in better employment, health, and quality of life outcomes
for SSDI beneficiaries. Answers to this question appear mainly in Chapter 4 (Outcomes) and
Chapter 8 (Utilization of Services). The second question requests an explanation of the
programmatic barriers to work, including a focus on why beneficiaries elected not to enroll in the
Chapter 1: Study Overview
1-3
MHTS when given the chance, why beneficiaries who did enroll did not engage in its services, and
why some beneficiaries who participated in the intervention were unsuccessful in their efforts to
return to work. The answers are complex and explored in many different analyses conducted
following data collection. Chapters 3-8 each offer answers to various parts of this question. The
third question focuses on interpretations of the primary outcomes of the study and the potential for
programmatic changes to support beneficiaries with schizophrenia or an affective disorder in their
efforts to return to work. The answers to this question focus on policy implications of the study
results, which appear in Chapter 9.
This final report includes a comprehensive description of the study, including the study design,
implementation of key intervention components, outcomes, and policy implications. The report
includes the nine chapters listed below.

Chapter 1: Background and Understanding, acquaints the reader with the significant policy
concerns that SSA has for this population, relates recent research that motivated the
study, and presents the key research questions.

Chapter 2: Study Design, presents the design features that drive the scientific and policy
objectives of the study.

Chapter 3: Enrollment and Participation presents analyses of the characteristics of SSDI
beneficiaries who enrolled in the study. The primary analyses include identification of
distinguishing characteristics of SSDI beneficiaries who enrolled compared to the larger
SSDI population who were eligible to participate in the study, as well as baseline
comparisons between beneficiaries randomized to the treatment and control groups.

Chapter 4: Outcomes, presents analyses of the key outcomes of the study and includes a
comprehensive analysis of the intervention’s impact on SSDI beneficiaries’ employment
and earnings (including months employed during the study, types of jobs obtained,
wages, etc.), health status, and quality of life.

Chapter 5: Implementation of Supported Employment and Other Behavioral Health and Related
Services, presents a description of the implementation strategy and fidelity of the SE and
OBH components of the intervention, and includes data on the actual implementation
parameters.

Chapter 6: Implementation of the Nurse Care Coordinator Role and Systematic Medication
Management (SMM), presents a description of the role of the NCC and SMM
intervention components, and includes data on the actual implementation parameters.

Chapter 7: Health Care and Supported Employment Financing, describes the approach used to
provide participants in the treatment group with access to medical, behavioral, and SE
services, and the resulting expenditures associated with providing those services.
Chapter 1: Study Overview
1-4

Chapter 8: Utilization of Services, presents a comparison of the use of behavioral health
care services (including emergency room visits and hospital overnight stays) between
participants in the treatment and control groups at baseline and during the study.

Chapter 9: Study Limitations, Key Findings, and Policy Implications, revisits the original
research questions posed by SSA, presents study limitations, concludes the findings of
the previous chapters, and offers a broad range of policy implications for SSA and other
federal agencies affecting provision of evidence-based services to SSDI beneficiaries
with severe mental illness.
Background and Understanding
SSDI is a long-term disability insurance program, managed by SSA, designed to provide income to
people who are no longer able to work because of a disability. Age at the time of onset of disability
determines the number of quarters a person must have previously worked in order to qualify for the
SSDI program. After a two-year waiting period, SSDI beneficiaries are eligible for Medicare benefits
provided by the Center for Medicare and Medicaid Services.
The SSDI program addresses the income support and health care needs of workers who can no
longer engage in substantial gainful activity (SGA) due to a medically determinable physical or
mental impairment. SGA is a formal term from the statute authorizing SSDI. SGA is set in
regulations and defines the monthly earnings level below which a beneficiary is eligible for cash
benefits (for 2011 the amount is $1,000/month or $1,640 for individuals who are legally blind).
However, any work activity can trigger a work CDR. It is possible that a beneficiary can lose
benefits because of either of these processes. Monthly earnings consistently above SGA will
eventually result in loss of cash benefits following both a grace period and a trial work period.
Regardless of work activity and level of earnings, the program requires a periodic medical CDR to
reestablish the existence of the underlying impairment that was the basis for disability or to
document medical improvement that can lead to program ineligibility. In cases where improvement
is expected, the next medical CDR occurs within 1 year. In cases where improvement is possible,
the next medical CDR occurs between 1 and 3 years. In cases where improvement is not expected,
the next medical CDR occurs within 5 to 7 years (SSA, 2010). In most cases of psychiatric
impairment, the CDR occurs within the 1 to 3 year range.
A high priority at SSA is supporting SSDI beneficiaries in their efforts to return to work. The agency
offers numerous programs and supports that facilitate return to work and pursuit of personal
employment goals. For example, the Ticket to Work program assists beneficiaries to get and keep a
Chapter 1: Study Overview
1-5
job. The Trial Work Period (TWP) allows beneficiaries to test their ability to work for weeks or
months at a time without worry of losing cash benefits. The Expedited Reinstatement rule serves as
a safety net for those beneficiaries who return to work, get off the program completely, but then
find at a later date that they can no longer work. Programs and supports like these exist to assist
beneficiaries in achieving their employment goals. The MHTS is one more in a series of ongoing
demonstrations at SSA to explore new and improved employment supports to beneficiaries in the
SSDI program. Other efforts include the Benefit Offset National Demonstration project and the
Youth Transition Demonstration (SSA, 2009).
SSDI Beneficiaries with Psychiatric Impairments
SSA categorizes individuals with psychiatric impairments as having a mental disorder in the SSDI
program. This classification also includes individuals with mental retardation. However, the majority
of SSDI beneficiaries with mental disorders are those with psychiatric impairments, including
schizophrenia, paranoia, and other psychotic disorders; affective disorders; anxiety-related disorders;
and personality disorders.
According to the SSA Listings for Mental Disorders, “The evaluation of disability on the basis of
mental disorders requires documentation of a medically determinable impairment(s), consideration
of the degree of limitation such impairment(s) may impose on the individual’s ability to work, and
consideration of whether these limitations have lasted or are expected to last for a continuous period
of at least 12 months” (SSA Blue Book, 2008, Section 12.00). Accordingly, consideration for SSDI
benefits first requires there be medical findings that substantiate the presence of a mental disorder.
Second, an assessment of the level of severity examines the impairment-related functional limitations
that are incompatible with the ability to engage in SGA. These include limitations related to activities
of daily living; social functioning; concentration, persistence, or pace; and episodes of
decompensation.
Beneficiaries with psychiatric impairments have been and continue to be a significant policy concern
for SSA. Since the early 1980s, the agency has struggled with regulations governing disability
determinations for this population (Grob & Goldman, 2006) and the reliability and validity of
standards for diagnosis of psychiatric impairments (Pincus et al., 1991). During this period, there has
been a steady, if not dramatic, growth in the number of SSDI awards for this population. In 1970,
program awards for psychiatric impairment were 2 percent of all new awards (Danziger, Frank, and
Meara, 2009). However, in 2006, the year MHTS implementation began new awards for this
Chapter 1: Study Overview
1-6
population that had increased to nearly 22 percent of all new awards (SSA, 2005). In addition, the
number of beneficiaries under the age of 50 continues to increase at a rate of nearly 3 percent per
year. Table 1-1 shows the number of SSDI beneficiaries under age 50 with “Other Mental
Disorders” (i.e., mental disorders other than mental retardation) between 1996 and 2009, along with
the number and percent change each year. In only one year (in 1997) did the number of SSDI
beneficiaries drop. The number of SSDI beneficiaries with Other Mental Disorders increased by
268,004 over the period from 1996 to 2009. This represented an increase of 38 percent (SSA, 2009).
Individuals classified as having a Mental Disorder are more costly than other SSDI beneficiary
populations, primarily because they are younger (in their mid 40s to mid 50s on average), have been
on the rolls on average for nearly a decade, and stay on the rolls longer (Hennessey & Dykacz, 1989;
McAlpine & Warner, 2000). In addition, one in four SSDI beneficiaries have income and resources
low enough that they also receive Supplemental Security Income (SSI) benefits from SSA
(Berkowitz, 2003; Newcomb, Payne, & Waid, 2003).
Table 1-1.
Number of SSDI beneficiaries under age 50 who had a primary diagnosis of Other
Mental Disorder1: 1996-2009
Year
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
1
Number
701,887
685,227
703,537
719,237
737,289
777,823
814,543
847,508
875,735
898,200
909,980
919,886
940,168
969,891
Change
Percent change
-16,660
18,310
15,700
18,052
40,534
36,720
32,965
28,227
22,465
11,780
9,906
20,282
29,723
-2.4
2.7
2.2
2.5
5.5
4.7
4.1
3.3
2.6
1.3
1.1
2.2
3.2
“Other Mental Disorders” includes schizophrenia and affective disorders, as well as organic mental disorders, anxiety related disorders,
somatoform disorders, personality disorders, substance addiction disorders, and autistic/other pervasive developmental disorders. The
category does not include mental retardation.
SOURCE: Annual Statistical Report on the Social Security Disability Insurance Program, 2009, Social Security Administration, Office of
Retirement and Disability Policy, Office of Research, Evaluation, and Statistics.
Once on the rolls, SSDI beneficiaries in general, and beneficiaries with Other Mental Disorders
more specifically, do not return to substantial work activity in any significant numbers. Using crosssectional statistics published by SSA in 2009 (Tables #6 and #52, SSA 2009), approximately eighttenths of one percent (.78%) of beneficiaries age 18 to 49 with Other Mental Disorders had their
Chapter 1: Study Overview
1-7
cash benefits withheld due to substantial work activity. An almost identical number (.76%) had their
benefits terminated in 2009 due to successful return to work (Tables #6 and #53, SSA, 2009).
Recent analyses conducted by Stapleton (2010), who used administrative records to follow a
longitudinal cohort of SSDI beneficiaries (with any diagnosis), suggested that benefit termination
due to work activity may be much higher, at around 4 percent. Regardless of whether one uses a
cross-sectional or longitudinal cohort, both percentages are remarkably low, suggesting that many
individuals who obtain SSDI benefits do not seek work and likely consider themselves as having left
the work force.
Use of SE to Improve Employment Outcomes
Research conducted in the past 20 years shows that many individuals with disabling mental illness
want to work and believe they could work if they had access to employment services and supports
(Frounfelker et al., in press; McQuilken et al., 2003; Mueser, Salyers, & Mueser, 2001; Uttaro &
Mechanic, 1994; Van Dongen, 1996). SE services, particularly the IPS model, have demonstrated
success in getting people with severe mental illness into competitive jobs. IPS is a well-defined form
of SE, and is an evidence-based practice specifically designed to serve individuals with severe mental
illness (Becker & Drake, 2003).
The body of evidence supporting IPS effectiveness began to develop in the early 1990s with quasiexperimental studies comparing day treatment programs to SE programs. When combined across all
of the early studies, comparisons between people receiving IPS modeled SE services to those
receiving day treatment showed that IPS was significantly more effective at increasing competitive
employment rates than the day treatment model.1 While 38 percent of the SE group achieved
competitive employment, the comparison group remained static at 15 percent competitive
employment (Bond, 2004).
Subsequent research employing the IPS model shifted to experimental trials with the goal of
establishing a causal relationship between IPS and employment. An empirical review of 11
randomized controlled trials of IPS programs serving individuals with severe mental illness
concluded that employment outcomes were consistently higher than the alternative control group
programs (Bond, Drake, & Becker, 2008). In these studies, participants in the control group received
1
Competitive employment is a job that pays at least minimum wage; is “owned” by the employer rather than a mental health center or rehabilitation
agency; not set aside for people with disabilities; and is carried out with some degree of regularity in terms of work schedule.
Chapter 1: Study Overview
1-8
either vocational services as usual, or a specific non-IPS vocational rehabilitation service. The length
of followup across the studies ranged from 6 to 24 months. The overall rate of competitive
employment was 61 percent for the participants receiving IPS services and 23 percent for
participants in the control groups. A subsequent meta-analysis (Bond, Xie, & Drake, 2007)
confirmed these findings in a similar sample of 681 individuals with a diagnosis of severe mental
illness that also varied in their receipt of SSI and SSDI.2
Despite this strong body of supporting evidence, the IPS model of SE is not widely implemented
across the United States, and is not available to many individuals with severe mental illness who
could potentially benefit from it. According to the Substance Abuse and Mental Health Services
Administration (SAMHSA, 2009), only about 2 percent of the population with severe mental illness
has access to SE services, and far fewer to the evidence-based IPS model. The reason is primarily
due to difficulties in financing these services (Drake et al., 2009). Local mental health service
providers cannot recover the costs associated with hiring, training, or retaining employment
specialists. In many states, Medicaid will pay for some portion of these employment services for
beneficiaries receiving SSI, but the limitations and complexity of accessing such coverage appears to
discourage even the most ardent supporters of these services. Similarly, state vocational
rehabilitation services cover some aspects of IPS services, but not others. Without reliable and
substantive financial support, it is not plausible for community mental health centers to provide SE
services to their clients. In a recent report to the Assistant Secretary for Planning and Evaluation in
the Department of Health and Human Services (DHHS), Karakus, Frey, Goldman, Fields, & Drake
(2010) addressed in detail the lack of access to evidence-based SE services for persons with severe
mental illness and the potential financing solutions to making those services more widely accessible.
Use of Medications to Improve Psychiatric Symptoms
Medications such as antipsychotics, mood stabilizers, and antidepressants provide symptom relief
for persons with schizophrenia, bipolar disorder, or depression (Hales et al., 2010). There is virtually
universal use of medication to treat these conditions in public and private mental health facilities in
the United States. In addition, prescribing antidepressants is extremely common among primary care
providers. Relatively recent studies, however, have identified significant problems with the ways in
which medications are used and with the adequacy of patient-based information needed to guide
2
It should be noted that all of these studies of the effectiveness of supported employment recruited participants from the rolls of community mental
health programs and not from the general rolls of beneficiaries on SSI or SSDI.
Chapter 1: Study Overview
1-9
individualized medication selection. The Schizophrenia Patient Outcomes Research Team (PORT)
study, for example, showed that both medication and psychosocial treatments for schizophrenia in
community settings are frequently not in accord with expert, research-based recommendations
(Lehman & Steinwachs, 1998). A number of studies identified underdosing of antidepressants as a
major issue in primary care settings. Chart reviews in public mental health clinics also found poor
documentation of schizophrenia symptoms more than half the time and an inadequate description
of side effects 85 percent of the time (Cradock et al., 2001).
Evidence from primary care and public mental health settings has demonstrated the feasibility and
effectiveness of a collaborative care model, in which a trained non-physician clinician works in
tandem with prescribers to document symptoms and side effects. Trained to recognize and evaluate
the disorders, symptoms, and side effects of particular medications, this non-physician clinician
prompts the prescriber with evidence-based recommendations based on the patient’s history and
characteristics, and the features of available medication treatments (Katon et al., 1995; Miller et al.,
2004; Suppes et al., 2003; Trivedi et al., 2004). Nurses with training in the use and side effects of
medications, and who regularly work directly with physicians, have the skill set necessary for
effective implementation of these collaborative models.
Disincentives for Returning to Work
Employment rates and the amount of work among people with severe mental illness are very low;
lower than those of people with other types of disabilities, and remarkably lower than those of
people without disabilities. Recent employment statistics released by the U.S. Bureau of Labor
Statistics (2011) indicate that in 2010 the labor force participation rate for people without disabilities
(ages 16 to 64) was 77 percent, while the employment rate was 70 percent. Comparable statistics in
2010 for people with disabilities in general suggest a participation rate of 34 percent and an
employment rate of 29 percent, both less than half the rates for the population without disabilities.
Research conducted by Mechanic, Bilder, McAlpine (2002) determined that the employment rate for
persons with severe mental illness is approximately 22 percent during any given period. Further, they
note that a substantial portion of the remaining population receives SSDI or SSI and does not
participate in the labor force. The U.S. Bureau of Labor Statistics (2010) reported that approximately
80 percent of workers without disabilities worked full time. The Mechanic, Bilder, and McAlpine
article noted that only about 12 percent of persons with severe mental illness work full time.
Chapter 1: Study Overview
1-10
For the majority of people with a severe mental illness, getting a job and working can be a struggle.
These individuals face a complex set of barriers that introduce doubt and confusion about their
ability to work. The barriers include the stigma and discrimination attached to mental illness;
disincentives unintentionally produced by the SSDI and SSI programs; and, as described above, the
inaccessibility of evidence-based behavioral health services and employment supports. In addition,
people with psychiatric impairments often have a relapsing-remitting course of illness that can
negatively influence work consistency and social relationships. While all of these barriers are
important, the barriers associated with the disincentives created by SSDI and SSI program
participation make the decision to return to work particularly complicated.
In a qualitative study of 16 employed and 16 not employed people with severe mental illness who
also received SSDI benefits, O’Day and Killeen (2002), found that the current federal policies and
practices associated with the program created disincentives for beneficiaries wanting to return to
work. Initially, the individuals studied found the SSDI application process to be difficult, timeconsuming, unpleasant, and even humiliating. Because of these difficulties in qualifying for benefits
and the ordeal they experienced during the application process, they were fearful of risking their
benefits, once obtained, to try to return to work. Even if they recovered and felt well enough to try
to return to work, there was fear of a relapse, resulting in anxiety about the prospect of having to reapply for benefits if symptoms returned. While this study included only small numbers of SSDI
beneficiaries, its rich detail provides important information not normally brought to light in studies
with larger populations.
The fear of losing Medicare or Medicaid that comes with SSDI or SSI benefits further creates
concerns about returning to work (MacDonald-Wilson et al., 2003; McQuilken et al., 2003; O’Day &
Killeen, 2002). MacDonald-Wilson and colleagues conducted a study of Social Security work
incentives and barriers to work. They surveyed 539 people with mental illness, 120 service providers,
and 174 family members. All three groups rated the potential loss of health insurance as the number
one barrier to returning to work. In another survey of consumers with mental illness receiving
services at an urban mental health center, McQuilken and colleagues (2003) reported a number of
reasons why SSDI beneficiaries did not return to work. Two key points that stood out were (1) the
fear of losing benefits if they returned to work, and (2) the difficulty of getting back on benefits if
they returned to work, and then lost their benefits. Three groups of beneficiaries consistently made
these points, including those who indicated that they did not want to work, those who indicated they
wanted to work but were not currently looking for work, and those who were looking for work.
Chapter 1: Study Overview
1-11
Even with successful return to work, health care receipt is an issue among individuals with severe
mental illness. Once on private health insurance, former beneficiaries will likely face limitations in
the amount of mental health treatment and type of medications that their new insurer will cover.
This prospect has improved with the passage of the “parity” legislation, removing differential costsharing and other insurance benefit limits between behavioral health care and general medical
services. Nonetheless, private health insurance does not cover some of the essential services that
support a return to work, in particular, SE services. The potential for trading Medicare benefits for
inadequate (or no) health care coverage only causes more hesitation in returning to work.
SSDI earnings limits are another major disincentive to returning to work. Beneficiaries report feeling
trapped in part-time, low-paying jobs even though they feel they are capable of earning more and
working in higher paying jobs. Some beneficiaries express a desire to work full-time, but know they
would lose their benefits under earning limits (Livermore, Goodman, & Wright, 2007; O’Day and
Killeen, 2002). Because of their education and skills, some of these beneficiaries are unlikely to earn
enough to make up for the SSDI benefits (including health insurance) they would lose by returning
to work full-time. The situation is particularly dire for SSDI beneficiaries who are also on SSI. For
SSI beneficiaries, after the first $85 of earned income, their SSI check is reduced by $1 for every $2
they earn. Many beneficiaries elect to forego jobs that place them in this position.
Summary of Background and Understanding
Many SSDI beneficiaries with severe mental illness want to work, and they can work, but they
struggle against difficult odds to get and keep jobs. As previously noted, it is relatively rare for these
beneficiaries to have access to the evidence-based treatments and supports necessary to improve
their functioning needed for work. While effective SE and mental health services exist, they are not
readily available to more than a small percentage of SSDI beneficiaries. However, research suggests
that, when available, these services improve employment rates. Whether or not access to needed
treatments and supports can overcome the strong disincentives resulting from SSDI program
participation is an unanswered question.
Design of the Mental Health Treatment Study
In response to the seemingly surmountable barriers to effective treatment and return to work, the
SSA designed the MHTS. The agency had high expectations that the study could demonstrate
Chapter 1: Study Overview
1-12
successful SSDI beneficiary access to evidence-based treatments and employment supports across a
representative sample of community mental health settings; and further, that such access would
result in improved employment rates, health, and quality of life.
Formal efforts to design the MHTS began more than a decade ago with an SSA Task Order request
to design an experimental study to test the effects of providing SSDI beneficiaries with affective
disorders access to SMM and SE services. In 1999, SSA announced the study as a historic
demonstration project that could help beneficiaries with affective disorders overcome the disabling
effects of their illness, return to the workforce, and lead more productive lives. At that time, 11
percent of the 4.7 million SSDI beneficiaries receiving disability payments had a primary diagnosis of
an affective disorder. Research during the previous decade strongly supported the need to test the
hypothesis that such access would lead to improvements in daily functioning and ultimately
employment. Conducted by The Lewin Group, the Task Order resulted in the release of a Request
for Proposals by SSA in June 2001, titled Affective Disorders Treatment Demonstration Project (RFP-010031, 2001). SSA accepted proposals, but later cancelled the procurement. After several attempts to
field the study, a second Task Order request, released in July 2003, called for a Technical Advisory
Panel to assist in the conceptual design of the MHTS (SSA-RFTOP-03-2006). The Task Order,
conducted by the Urban Institute, released its final report in February 2005 (Aron, Burt, &
Wittenberg, 2005). The report recommended expansion of the study to include individuals with
schizophrenia. In May 2005, after more than 5 years of development, the SSA released a request for
proposal (RFP) to conduct the MHTS (SSA-RFP-05-1044).
SSA’s design for an effective test of access to SE and SMM services, in the face of an emerging
reality that many SSDI beneficiaries want to work and can work, is a reflection of the complexity of
SSA disability policies. At present, being on the rolls and working are effectively (though not
technically) mutually exclusive. Beneficiaries that want to work are encouraged to work by SSA. In
fact, SSA offers a variety of programs to encourage work (SSA, 2009; Livermore, Goodman, &
Wright, 2007; MacDonald-Wilson et al., 2003). On the other hand, SSDI beneficiaries who do work
face reduced cash payments if they work “too much.” If they work at or above SGA for lengthy
periods, they also risk losing their medical benefits due to program ineligibility. The medical benefits,
while not necessarily sufficient, are still an important safety net for those who fear recurrence of
mental illness symptoms.
Chapter 1: Study Overview
1-13
Outcome Measures
SSA specifically targeted three outcome areas of study for beneficiaries participating in the MHTS.
These areas included employment, health, and quality of life. While there were many other areas of
secondary interest, these three areas were the primary focus of the agency in its attempt to
understand the impact of the intervention on SSDI beneficiaries with schizophrenia or an affective
disorder.
Employment. The primary employment measure was employment rate, as measured by beneficiary
self-report of work at any point during the study period. Beneficiaries responded to standardized
questions about employment (Current Population Survey, sponsored by the Bureau of Labor
Statistics, 1994-present) during each of nine interviews spaced approximately every quarter
throughout the 24 months of study participation.
Health. The health measures were physical and mental health status based on beneficiary selfreports of perceptions of their health during the previous month. Assessed at baseline and again at
study exit, the key measures were the SF-12 component scores that summarize overall physical and
mental health (Ware, Kosinkski, Turner-Bowker, and Gandek, 2002).
Quality of Life. The measure of quality of life was a single self-report scale item from the Modified
Lehman Quality of Life Inventory (QOLI-M). The QOLI-M is a shortened version of the wellknown QOLI (Lehman, 1988). Similar to the health status measures described above, the quality of
life measure was assessed at baseline and study exit.
Secondary Outcomes of Interest. The four measures mentioned above—employment rate,
physical health status, mental health status, and quality of life—form the heart of the outcome
analysis presented in Chapter 4. However, the study offered the opportunity to assess the
effectiveness of the intervention on a number of additional outcomes of secondary interest to SSA.
One set of these outcomes concerned variables related to employment, including earnings, job type,
hours worked, highest hourly wage, and satisfaction with job, among others. Additional outcomes of
secondary interest were income from a variety of sources (e.g., other public programs, investments,
savings, alimony, etc.), alcohol and substance abuse, and health care (including mental health care)
service utilization levels and patterns.
Lastly, there was interest in the outcomes of some specific subgroups of beneficiaries participating in
the study. During discussions about sample size for the study and statistical power, SSA noted
Chapter 1: Study Overview
1-14
potential interest in diagnosis, age, and gender, and education. Accordingly, these variables also
received attention in the outcome analysis, though they were secondary to the overall impact of the
intervention on all study participants.
MHTS Investigators
Westat was the prime contractor for the MHTS. Supporting Westat were researchers from
Dartmouth Medical School, the University of Texas Health Sciences Center at San Antonio, and the
University of Maryland Baltimore County.
Co-Principal Investigators
William Frey, Ph.D., from Westat, and Robert Drake, M.D., Ph.D., from Dartmouth Medical School
served as Co-Principal Investigators of the study. Dr. Frey, a vice president and study area director
in the Health Studies sector at Westat, provided overall management for the study. Dr. Drake, who
directs the New Hampshire-Dartmouth Psychiatric Research Center, provided support for treatment
implementation at the study sites.
Westat
Led by Dr. Frey, the Westat team developed the study protocols, oversaw the day-to-day operations,
and served as the primary point of contact for the 23 study sites. The Westat team also oversaw the
management of the MHTS beneficiary sample, designed and implemented the study management
information systems and computer-assisted personal interviews (CAPI) used for the study, and led
the data cleaning and analytic file preparation efforts. Staff also played a key role in the data analysis.
Howard Goldman, M.D., Ph.D., a Westat consultant and Professor of Psychiatry at the University
of Maryland, School of Medicine, served as mental health policy expert to the study. An expert on
mental health disability policy, Dr. Goldman provided the study team with ongoing consultation
regarding all aspects of the MHTS.
Chapter 1: Study Overview
1-15
Dartmouth Medical School
Robert Drake, M.D., Ph.D., Gary Bond, Ph.D., and Deborah Becker, M.S., provided
implementation support, fidelity measurement, and quality assurance associated with the provision
of SE and OBH services to study participants at the 23 study sites. They also provided oversight
through a group called the Quality Management Program Directors who provided ongoing technical
assistance to the study sites and conducted on-site SE fidelity visits.
The University of Texas Health Science Center at San Antonio
Alexander Miller, M.D. and Troy Moore, Pharm.D, M.S. Pharm., BCCP with the University of
Texas Health Science Center at San Antonio provided implementation support, fidelity
measurement, and quality assurance associated with the provision of SMM services to study
participants at the 23 study sites. Drs. Miller and Moore also conducted site visits to the 23 study
sites to ensure that NCCs followed the SMM intervention protocol for treatment group participants
who were prescribed psychiatric medication.
The University of Maryland, Baltimore County
David Salkever, Ph.D., a senior health economist and leading expert in the evaluation of mental
health treatment and economics of disability, led the impact analysis of the study outcomes.
Chapter 1: Study Overview
1-16
Chapter 2
Study Design
The Social Security Administration (SSA) developed the research agenda for the Mental Health
Treatment Study (MHTS) with guidance from its Technical Advisory Panel (TAP) (Aron et al.,
2005). The final study design is the focus of this chapter. It incorporates the priorities of SSA,
recommendations of the TAP, as well as insight and contributions from the investigators. The
chapter begins with an overview of the two predominant scientific features of the study design—the
experimental nature of the design and the intent-to-treat (ITT) approach to data analysis. These
features were central to obtaining a clear assessment of the impact of the intervention on study
participants’ employment, health, and functioning. The next section addresses the target population,
and includes descriptions of study site selection, sampling methodology, and the randomization
process. The next sections describe the intervention approach, including the intervention benefits
and services provided to the treatment group, the control condition, and the quality control
(fidelity) measures used to monitor implementation of the intervention. The concluding sections
describe the outcome measures, data collection procedures, data preparation, statistical procedures
for analysis of the results, and data limitations.
Key Scientific Features
The study design provides SSA with an objective and unbiased assessment of the impact of
providing the intervention package to its SSDI beneficiaries who enrolled in the study. The two
predominant features of the study design were its randomized controlled trial (RCT) structure and
the ITT principle of the data analysis. Both are features of a strong experimental study design that
contribute to arriving at answers to the primary research question about the impact of the
intervention on employment, health, and quality of life.
RCT Design
The primary intervention components of the study—systematic medication management
(SMM) and the Individual Placement and Support (IPS) model of supported employment (SE)
services—have strong evidence to suggest their efficacy. Researchers have demonstrated success
Chapter 2: Study Design
2-1
with each component in previous studies (including randomized trials) with a variety of populations
of individuals with mental illness in a variety of clinical settings. However, their effectiveness
remained untested with the SSDI population exclusively, in combination with one another, and
across the diverse mental health treatment settings that reflect the existing service structures in the
United States. The MHTS tested the intervention components under these circumstances with a
two-arm RCT design. The intervention components (treatment group) comprised one arm of the
design, and a services as usual (control group) comprised the second arm.
The primary value of the randomized design was reduction in allocation bias provided by random
assignment of SSDI beneficiaries to either the treatment group or the control group. Doing this
improved the chances that treatment and control group participants were as similar as possible at the
outset of the study, especially with regard to baseline measures of employment, health, and
functioning. If participants proved to be different on any of these baseline measures, then additional
design or analysis procedures would have been necessary. Following random assignment, tests of
equivalence between the treatment and control groups on these primary baseline measures
confirmed the success of the randomization process. Details of the randomization process appear
below. The test results appear in Chapter 3 Enrollment and Participation.
ITT Principle
Another study design feature of the MHTS was the ITT principle. ITT is fundamental to RCTs.
Data analysis includes all randomized subjects according to the group to which they were allocated
(Lachin, 2000; Hollis & Campbell, 1999). Without this principle, even randomized trials may be
subject to bias. For example, in the MHTS, participants in the treatment group could have
withdrawn for any number of reasons, leaving only those who were best able to look for and obtain
a job. If the study were to evaluate only the outcomes of those who remained in the study from
beginning to end, and ignore the withdrawals, then the outcome would not represent the population
of all beneficiaries enrolling into the study. For these reasons, the MHTS followed the ITT principle
so that the outcomes for all study participants would be included in the analysis regardless of
whether they withdrew from the study, the extent of adherence to the treatment intervention, or
their level of engagement.
SSDI beneficiaries randomized to the treatment group were not required to participate in all study
activities in order to remain in the study. For example, if they elected not to take prescription
medications, then they were less likely to participate in the SMM component of the intervention. In
Chapter 2: Study Design
2-2
addition, they were not required to see a SE specialist at the study site. Designers of the study
anticipated that some SSDI beneficiaries with mental illness would be unable or unwilling to
participate in certain study activities, would become hospitalized, or refuse certain intervention
components. Other beneficiaries might completely disengage from the study after enrollment, no
longer participating in any study activities. The study did not “drop” these partially engaged or
unengaged participants. Instead, study procedures required local study site staff to attempt to reengage these beneficiaries throughout the period in which they were participants. Study participants
in either the treatment group or control group could voluntarily withdraw from the study in which
case they received no further study contacts. However, the analysis included data available before
such participants voluntarily withdrew from the study.
The study design included two strategies to account for beneficiaries who withdrew or otherwise
had missing data at study end. If an enrollee did not complete at least two post-baseline self-report
interviews, then these cases were non-respondents. Weights were calculated and applied to the
remaining study population to adjust for this type of non-response. Some enrollees withdrew later or
did not complete all research interviews. In these cases, imputed data replaced the missing data. A
more detailed discussion of each strategy appears later in this chapter in the section titled Data
Preparation and Adjustments.
Study Site Selection
Two of SSA’s priorities were to implement the study in real world mental health settings and to do
so in all regions of the United States. These priorities presented the investigators with particularly
difficult challenges, given research clearly indicated that the core services were not readily available
in most communities throughout the United States. Thus, it was a challenge for the investigators to
identify community mental health centers with the capacity to provide the primary intervention
components and then prepare them for delivering the necessary services.
The TAP convened by the Urban Institute also noted that the MHTS should cover multiple states
supporting the goal that the demonstration be representative of the types of community mental
health centers found in the United States. One potential strategy for achieving representation was
random selection through a nationally representative probability sample of study sites. However, the
investigators decided against this strategy for two reasons. First, the study schedule required a rapid
start-up with sites fully functioning within 6 months. It would have been impractical to randomly
select study sites, ensure the necessary services were in place, and have the key components of the
Chapter 2: Study Design
2-3
intervention—IPS SE services and SMM services—fully operating at high fidelity to the evidencebased practice within this time frame. Second, the number of study sites to be selected (originally
20) constituted a rather small number for a national probability sample. In addition, with the random
selection of study sites approach, the level of fidelity would likely have been highly variable, leading
to potentially ambiguous results. For example, it was possible that the results would not be clear as
to whether the intervention was ineffective, or the ability of the sites to deliver high fidelity services
was limited. SSA and the investigators concluded that the sites needed to deliver high fidelity
services in order to establish the effectiveness of the treatment intervention with the target
population.
The investigators selected the study sites from among the more than fifty IPS programs that were
already in operation, while taking additional steps to ensure the selection of sites was as diverse as
possible and reflected various types of community mental health centers in the United States. In
selecting the sites, considerations included U.S. census region, urbanicity, population composition,
and center organization. Most of the selected sites had a history of delivering SE services, although
not necessarily at high fidelity to IPS. With one exception (described below), they all delivered
medication management services and were willing to participate in SMM.
Initially, the selection process resulted in a recommendation of 20 study sites to SSA. SSA approved
all 20 sites but requested recommendations for two additional study sites to ensure Hispanic
population representation in the study. The sites identified were two mental health centers with
strong interest in SE services in New York City, New York and San Antonio, Texas. Both sites
served large Hispanic populations. While neither had existing IPS SE, both centers had experience
with the provision of other forms of SE services and were willing to hire staff and undergo rapid
training in order to provide IPS services. Upon approval, both of these additional sites received
extra resources and training in advance of study startup to enhance their capacity to implement IPS
SE.
The study added an additional (twenty-third) study site approximately one-third of the way through
the recruitment period. The inability of the original south Chicago site to handle potential study
participants in the northern part of the catchment area necessitated the addition of a new site on the
north side of Chicago. Operated by the same community mental health agency, this site received
training and startup support from the south Chicago site.
The map in Figure 2-1 shows the location of the twenty-three sites across the United States. One of
the selected study sites (Spring Lake Park, Minnesota) was chosen to increase geographic diversity by
Chapter 2: Study Design
2-4
its location in the northern Midwest region. Known for its excellent vocational services, this agency
delivered SE services directly to enrollees, but not behavioral health and related services. This site
had an existing contractual relationship with a county agency to provide these services. Discussions
that occurred during the site selection process resulted in the understanding that such situations
were a natural part of the continuum of community mental health service arrangements available
across the country, and therefore, were an appropriate addition to the study.
Figure 2-1.
Map of MHTS study sites
Two study sites faced internal operational issues within the first year of the study, which negatively
influenced their ability to continue with the study. As a result, these sites ceased recruitment and
enrollment activities within the first year (in June and September 2007). One site reorganized with
involvement and oversight from their state office of mental health. The upheaval created by the
reorganization was too disruptive to study operations to continue enrollment. The second site
reported that their indirect costs were too high to continue participation in the study, and thus
requested to withdraw. Both study sites terminated the two research positions of Nurse Care
Coordinator (NCC) and Research Assistant (RA) but agreed to continue to provide treatment and
SE services to the small number of beneficiaries already enrolled in the study. Treatment and control
Chapter 2: Study Design
2-5
group participants at these sites also continued to complete the post-baseline self-report interviews
throughout their 24-month period of participation. Their study data were included in the analysis
along with that of participants from all of the study sites. Most of the other sites faced severe
financial stress due to the economic recession and state funding cutbacks, but were, nevertheless,
able to sustain the treatment intervention.
Sampling Methodology
The investigators identified a target population of SSDI beneficiaries residing within the catchment
areas of the study sites, organized the beneficiaries into relevant strata, and then randomly assigned
beneficiaries (proportional to each stratum) to recruitment release groups with a size of 25
beneficiaries per release group. Research staff at each study site then proceeded to recruit SSDI
beneficiaries from their release groups.
The target population for the study included SSDI beneficiaries between the ages of 18 and 55
inclusive with a primary diagnosis of schizophrenia or an affective disorder. With the few exceptions
noted below, SSDI beneficiaries living within one of the study site catchment areas who met these
criteria comprised the study target population. The exceptions were those beneficiaries living within
the study site catchment area with one or more of the following situations:

Beneficiaries living in a nursing home or other custodial institution;

Beneficiaries designated legally incompetent to manage their own affairs (i.e., had a legal
guardian);

Beneficiaries with a life-threatening or a terminal condition (e.g., terminal cancer, AIDS,
or end-stage renal disease 1);

Beneficiaries receiving SE services from their participating study site within six months
of study enrollment; and

Beneficiaries with a competitive job less than thirty days prior to study enrollment.
Define primary and backup catchment areas. The catchment area refers to the specific postal
zip codes included in the communities, cities, or counties that the study site typically serves. With
1
Prior to completing the Baseline interview, beneficiaries were asked if they had any diseases, disorders, or physical impairments that would prevent
them from working, receiving SE services, or participating in any other study activities. If a beneficiary reported having terminal cancer, AIDS, or
end-stage renal disease, he or she would have been excluded from the MHTS; however, no beneficiaries met this exclusion criteria (i.e., no
beneficiaries reported having terminal cancer, AIDS, or end-stage renal disease prior to completing the Baseline interview).
Chapter 2: Study Design
2-6
the relevant zip codes identified, the administrative records from the SSA Master Beneficiary Record
(MBR) file permitted identification of potentially eligible beneficiaries. The investigators held
meetings with each study site to define their primary catchment area by geography (community, city,
or county) and zip code. When appropriate, the primary catchment area was based on a combination
of the service area from which the site was legally obligated (allowed) to draw clientele, as well as the
estimated distance potential beneficiaries might be willing to travel to receive services from the site.
The defined catchment area included an approximate 30-mile radius around the study site; however,
the service area of the site took precedence over distance from the mental health center. Each study
site also identified zip codes for a “backup” catchment area from which to recruit if a sufficient
number of beneficiaries did not enroll from the primary catchment area. Since the study sites
defined the backup catchment areas, the composition of these areas varied but included a
combination of (1) beneficiaries who lived outside of the approximate 30-mile radius around the
study site and (2) beneficiaries who lived in an adjoining county.
Identify sample for target population. SSA provided the investigators with an electronic file from
the MBR containing the target population with eligible SSDI beneficiaries. The electronic file
contained a number of relevant variables needed for sampling, such as name and contact
information, Supplemental Security Income (SSI) status, date of birth, and diagnosis. Table 2-1
presents a summary of the sample for the target population for each study site by psychiatric
impairment type (diagnosis) and gender. The table contains two categories of SSDI beneficiaries—
beneficiaries receiving SSDI only and beneficiaries receiving both SSDI and SSI (i.e., dual eligible
beneficiaries). The ratio of beneficiaries with an affective disorder to schizophrenia is 2 to 1 (68% to
32%). While the overall ratio of males to females is slightly smaller than 1 to 1 (48% to 52%), the
percentage of males to females among beneficiaries with schizophrenia is about 2 to 1 (67% to
33%). However, substantially more beneficiaries with an affective disorder are female; the ratio of
females to males among beneficiaries with an affective disorder is approximately 1.6 to 1 (61% to
39%).
Apply pre-recruitment exclusion criteria. It was possible to exclude some SSDI beneficiaries
living within the catchment areas prior to release of the sample to the study site. The MBR
contained variables that identified individuals as having a legal guardian, or living in a nursing home
or other custodial institution. Thus, the final sample of eligible beneficiaries did not include those
with such designations. In order to eliminate beneficiaries whose MBR data were not up to date, the
screener that preceded the Baseline interview also included questions to identify these individuals
and remove them from the pool of eligible beneficiaries.
Chapter 2: Study Design
2-7
Table 2-1.
MHTS sample by key stratifications
Interview
SSDI
Male
18-35
36-55
Female
18-35
36-55
<24 months on rolls
≥24 months on rolls
SSDI with concurrent SSI
Male
18-35
36-55
Female
18-35
36-55
Schizophrenia
(N=19,822)
freq
%
14,055
70.9
9,537
48.1
1,749
8.8
7,788
39.3
4,518
22.8
588
3.0
3,930
19.8
690
3.5
13,365
67.4
5,767
3,728
1,081
2,647
2,039
404
1,635
29.1
18.8
5.5
13.4
10.3
2.0
8.2
Affective Disorder
(N=41,708)
freq
%
32,909
78.9
13,200
31.6
1,641
3.9
11,559
27.7
19,709
47.3
2,506
6.0
17,203
41.2
2,911
7.0
29,998
71.9
8,799
2,882
794
2,088
5,917
1,345
4,572
21.1
6.9
1.9
5.0
14.2
3.2
11.0
Total
(N=61,530)
freq
%
46,964
76.3
22,737
37.0
3,390
5.5
19,347
31.4
24,227
39.4
3,094
5.0
21,133
34.3
3,601
5.9
43,363
70.5
14,566
6,610
1,875
4,735
7,956
1,749
6,207
23.7
10.7
3.0
7.7
12.9
2.8
10.1
Three additional exclusion criteria could not be determined from the administrative data. One
criterion, current competitive employment, was determined from a series of employment questions
asked during the recruitment process. Beneficiaries competitively employed within the past thirty
days were not eligible for the study. A second criterion, existence of a physical health condition that
precluded study participation, was also determined from the screener, or through the General
Medical Exam (GME) administered to treatment group participants following random assignment.
During the screener, beneficiaries responded to a question that asked whether they had any diseases,
disorders, or physical impairments that would prevent them from working, receiving SE services, or
participating in any other study activities. The beneficiary was ineligible if he or she responded “yes”
and then reported having terminal cancer, AIDS (being HIV positive was not on its own grounds
for ineligibility), or end-stage renal disease. For participants randomized to the treatment group, their
primary care provider assessed all other identified conditions to determine study eligibility.
Beneficiaries who had received SE services from a study site within the past six months were not
eligible for the study. As the study site began recruitment efforts, a comparison between the list of
potentially eligible beneficiaries and the client list of the study site yielded a few beneficiaries who
received services within the past six months. Consequently, these beneficiaries were declared
ineligible and removed from the recruitment list.
Chapter 2: Study Design
2-8
Finally, the initial sample of eligible beneficiaries did not include those who were in the SSDI
program for less than 24 months. This group of beneficiaries was of interest to SSA. However, they
did not have access to Medicare until their 24th month of program eligibility, and there was concern
that they would pose a financial risk to the study. Once it was clear that study funding could
accommodate these individuals, the study sites received their names and contact information. The
inclusion of these individuals occurred near the end of the first year of recruitment. Treatment
group participants residing in Connecticut and Maryland who were in the 24-month waiting period
were recruited from the beginning of the study since these states offered access to health insurance
without precondition (see Chapter 7 Health Care and Supported Employment Financing for further details).
Release sample for recruitment. The final step of the sample specification process involved
formatting and loading the eligible target population of SSDI beneficiaries into an electronic Study
Management System (SMS). Organization of the sample for each study site included multiple release
groups in blocks of 25. Each release group contained the names and contact information of 25
beneficiaries selected from a stratified sort of the final list of eligible beneficiaries. The study sites
received access to two release groups at a time through the SMS. The release groups were activated
(i.e., made available to the study site) on a weekly basis or as needed by the site. The number of
groups released each week varied over time based on the site’s recruitment progress.
The primary reason for segmenting the release groups was to ensure that sites would “work” the
sample of beneficiaries and not “speed-call” all of the beneficiaries in the sample to identify those
who would readily enroll in the study. The philosophy was to make a good faith attempt to offer the
study to all eligible beneficiaries in the study site catchment area. A detailed description of the
recruitment process and information provided to potential beneficiaries appears in Supplemental
Appendix A, Study Implementation (see Recruitment section).
Sample Size
The initial sample size was 3,000 participating beneficiaries, with a target of 1,500 in each study arm.
However, a preliminary review of early study data prompted a downward revision of this initial
target. Estimates for both overall and subgroup comparisons based on data collected from the first
three months of the study indicated that a sample of 2,000 participants would provide sufficient
power to address the primary questions of interest for overall comparisons between treatment and
Chapter 2: Study Design
2-9
control groups, as well as many of the subgroup analyses. Specifically, this sample size would be
more than adequate (80% power or greater) to achieve the following:
1.
Find a $0.50 per hour wage difference in the overall sample;
2.
Show an increased proportion of persons working in the treatment group for
beneficiaries with schizophrenia, affective disorder, or aged 36 to 55;
3.
Determine whether the treatment had a beneficial effect on total monthly earnings; and
4.
Detect differences in the number of hours worked among persons with affective
disorder and persons aged 36 or older.
While the original sample size of 3,000 would provide enough power to detect differences in all of
the above variables and subgroups, it would not provide enough power to evaluate most outcomes
in the smaller subgroups, e.g., younger beneficiaries, beneficiaries with schizophrenia, or
beneficiaries with the shortest time receiving SSDI benefits. Based on the power calculations and
extensive discussions with SSA, the target population size decreased from 3,000 to 2,000.
The overall final sample size for the study was 2,238 SSDI beneficiaries, with 1,121 allocated to the
treatment arm of the study and 1,117 to the control arm of the study. The increase occurred for two
reasons. As the end of recruitment neared, SSA requested an additional 100 beneficiaries for the
treatment group to ensure that the numbers would be large enough for planned comparisons of
subgroups. Thus, approximately 200 additional beneficiaries were required in order to meet this
request due to the randomization procedures. Second, all study sites active in enrolling beneficiaries
at the end of the recruitment period had an end date for enrollment of July 31, 2008. As the final
days neared, it was not possible to stop enrollment precisely at 2,200 for a variety of reasons. Some
sites had prior commitments to beneficiaries, and in other cases, appointment dates for enrollment
already existed.
Overview of the Enrollment and Randomization Process
Enrollment into the study and random assignment to the treatment or control group followed a
structured and lengthy process. Research staff at each study site received extensive training on a
standard set of recruitment procedures designed specifically for the study. A detailed description of
these enrollment procedures appears in Supplemental Appendix A, Study Implementation.
Chapter 2: Study Design
2-10
Several steps preceded beneficiary enrollment into the study. Beneficiaries in each release group
received a letter introducing the study. Study staff subsequently contacted the beneficiaries by
telephone to invite them to attend a Recruitment Information Group (RIG) meeting. The RIG
meetings provided beneficiaries with complete information about the study as well as the
opportunity to have their questions answered. The use of RIG meetings is now standard practice in
RCTs of IPS and justification for its use in the psychiatric population well described (Drake et al.,
1994). Two RIG meetings were required before a beneficiary could enroll in the study. When a
beneficiary decided to enroll in the study, he or she first completed a competency screener (see
Supplemental Appendix B, CAPI Screener) to ensure the ability to provide informed consent. 2 Following
the competency determination, the enrollment process involved obtaining written consent to
participate in the study, and participating in a Baseline interview.
At the end of the Baseline interview, the site RA completing the interview called into Westat’s
WesTrax™ system, an automated clinical trial management tool, to obtain the randomization
assignment. The RA entered identifying information for him/herself and the beneficiary, and the
system responded with a treatment or control group assignment for the beneficiary. The RA
informed the beneficiary of the decision and proceeded to close out the interview and discuss next
steps with the beneficiary depending upon the randomization assignment.
The randomization scheme developed for the study made assignments in equal proportions (1 to
1) using a stratified permuted block randomization with a fixed size of four. The latter ensured
balance between the two study arms within the four identified strata, including (1) beneficiary on
SSDI < 24 months and not receiving SSI, (2) beneficiary on SSDI ≥ 24 months and not receiving
SSI, (3) beneficiary receiving SSI, and (4) study site. For example, with the two treatment arms, a
permuted block randomization with block size of four could yield the following allocations: ABAB;
ABBA; BAAB; or BABA.
2
No beneficiaries failed the competency screener.
Chapter 2: Study Design
2-11
Intervention Benefits and Services
The TAP identified seven primary characteristics of the MHTS intervention. Listed in the technical
report (p. 14, Aron, et al., 2005), these characteristics included the following:
1.
The primary focus of all aspects of the intervention (including clinical services and
supports) should be on promoting work.
2.
Great care should be taken to (a) establish an accurate medical diagnosis (including
secondary and/or confounding physical/mental medical conditions), (b) follow wellestablished treatment guidelines for the given diagnosis, (c) attend to the effects of
symptoms/impairments on executive and cognitive functioning rather than medical
conditions/diagnoses.
3.
The MHTS must adhere to established supported employment principles, specifically as
exemplified in the IPS approach.
4.
Clinical and employment support must be fully integrated with one another (one set
cannot be offered without the other).
5.
The intervention should rely primarily on established “evidence-based practices.”
6.
All communities participating in the MHTS must serve the same target populations and
adopt the same general approach (in other words, they cannot limit their intervention to
only one diagnosis, to SSDI applicants only, etc.).
7.
Given the challenges likely to be involved in recruiting participants, an important part of
the intervention will be an education and training component during an orientation
meeting. This meeting could also be used to obtain informed consent from participants
and include any additional screening for the target population.
The seven characteristics identified above effectively define what SSA meant when it referred to
providing SSDI beneficiaries with “access” to evidence-based services. When made operational, the
study defined access to services by the study benefits and services described in the paragraphs
below. Beneficiaries assigned to the treatment group in each of the study sites received these
benefits and services to the extent possible.
Diagnostic psychiatric assessment. As recommended by the TAP and required by SSA, the first
step in providing full access to care consisted of establishing an accurate and current psychiatric
diagnosis for each beneficiary. The diagnostic assessment, typically conducted within the first month
of study participation, used the Structured Clinical Interview for DSM-IV (SCID) Axis I Disorders.
The 4th edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) organizes all
psychiatric disorders and other problems into five different categories (or axes), however as
Chapter 2: Study Design
2-12
mentioned, the SCID was used to assess the presence of Axis I disorders only for MHTS
participants randomized to the treatment group. Examples of Axis I disorders include:

Anxiety disorders (e.g., panic disorder, social anxiety disorder, posttraumatic stress
disorder);

Mood disorders (e.g., major depression, bipolar disorder);

Eating disorders (e.g., anorexia nervosa, bulimia nervosa);

Psychotic disorders (e.g. schizophrenia and schizoaffective disorder);

Dissociative disorders; and

Substance use disorders.
Clinical psychologists in each study site received special training on conducting the SCID.
General Medical Exam (GME). The GME was an SSA requirement for participation in the
treatment group. The GME confirmed the physical fitness of the beneficiary for work. Specifically,
the GME policy required a review of recent and past health history and a brief physical examination
to ensure that working would not pose harm to the beneficiary. The GME requirement was
described during RIG meetings, as a part of the consent process, and at enrollment. Additionally,
once a participant was randomized to the treatment group, the GME requirement was again
explained during the post-randomization meetings. Failure of the GME or refusal 3 to complete a
GME prompted dismissal from the study. The study treated these beneficiaries as non-respondents.
These individuals were included in the analysis through weights created to represent their
participation in the study. Forty-two beneficiaries in the treatment group became administrative
drops due to a failed GME (8) or refusal to complete the GME (34).
SMM services. SMM for participants in the treatment intervention consisted of two parts. First, a
nurse experienced in working with mentally ill clients (referred to as the NCC) provided a systematic
assessment of participant symptoms and functioning to the physician prior to each prescriber visit,
or quarterly (whichever period was shorter). These assessments provided a structured view of the
participant’s overall response to current medication and cognitive functioning. The second
component included an algorithmic approach to prescribing psychiatric medications. The SMM
component of the intervention included the Texas Medication Algorithm Project
3
With minimal exception, those treatment group participants refusing to complete the GME were beneficiaries who had also failed to engage in the
SMM or IPS services and tended to be non-responsive to outreach attempts by site staff.
Chapter 2: Study Design
2-13
(TMAP) medication algorithms regarding schizophrenia and mood disorders. The TAP
recommended TMAP as a “rigorous demonstration of the implementation of step-wise decision
trees for specific psychiatric diagnoses in public mental health treatment centers” (Aron et al., 2005).
In addition to the algorithms, TMAP includes recommendations about dosage, length of treatment,
use of concomitant medications, evaluation of non-responders, use of rating scales, and other
medication issues.
SE services. In line with the TAP recommendation for SE services, the lead criterion for selecting
a study site was its demonstrated provision of high fidelity IPS services. Each study site was required
to have the infrastructure in place to provide IPS SE services with a ratio of one SE specialist to
every 25 beneficiaries.
Other behavioral health (OBH) and related services. The intervention package also provided
access to evidenced-based behavioral health services over the course of a beneficiary’s enrollment.
OBH and related services included but were not limited to psychosocial services, medical services,
substance abuse services, housing, legal services, family intervention, financial services, case
management, and benefits counseling. While desirable, the ability to provide all of these services was
not a requirement for study sites. However, the expectation remained that the study sites provide
most or all of these services.
Health insurance coverage. The study reimbursed treatment group participants for health
insurance premiums over the course of their participation in the study. The majority of MHTS
participants had Medicare, Medicaid, private insurance, military insurance, or some form of statelevel insurance. Specifically, the study paid each participant’s portion of his or her monthly health
insurance premium as well as the participant’s co-pays for behavioral health-related services. The
study provided an insurance coordinator to assist with any changes in Medicare Part D prescription
coverage plans to control costs and to ensure the most appropriate plan for each participant’s
medication profile.
Supplemental health insurance. A goal of the treatment intervention was to ensure that all
participants had some form of health insurance coverage that was comparable to Medicare Parts A,
B, and D. When supplemental health insurance was required, the study paid the premiums. There
were two primary reasons why a participant required supplemental health insurance—either the
participant was uninsured at enrollment, or the participant had Medicare, but only Part A. Those
participants without health insurance at enrollment were in the 24-month waiting period for
Medicare. However, not all participants in the 24-month waiting period were without health
Chapter 2: Study Design
2-14
insurance. Some had coverage under plans of other family members. Further discussion of
supplemental health insurance appears in Chapter 7 Health Care and Supported Employment Financing.
Coordination and payment of participant out-of-pocket and other non-covered expenses. At
times, treatment group participants had to pay for services that the study could not pre-pay. In these
cases, the participant was required to submit expenses and request reimbursement. In addition, there
were expenses not covered by health insurance or by other means. Some examples were studyapproved work-related expenses (e.g., license fees, special work clothes, tuition for special classes),
and the cost of transportation to and from providers for treatment participants who would
otherwise not have access to care.
Criterion number four of the TAP report states that “clinical and employment services must be fully
integrated with one another.” Implementation of this criterion was a goal of the study. However, the
extent to which this requirement was possible is subject to question. Early in the enrollment process,
it became clear that many beneficiaries who wanted to work and enroll in the study also wanted to
keep their current (off-site) clinical provider. SSA emphatically supported the views of these
beneficiaries. Thus, the study promulgated a policy to that effect and enacted procedures to improve
coordination and service integration between on-site and off-site providers. While co-location is not
a requirement for full integration of clinical and SE services, it clearly improves the probability for
such integration. Further discussions of the issue of on-site versus off-site providers appear in the
study implementation chapters (Chapters 5 and 6).
Suspend medical Continuing Disability Reviews (CDR). Participation in the study included
suspension of the beneficiary’s medical CDR for 36 months from the date of enrollment into the
MHTS. The goal of the CDR suspension was to encourage work activity and to remove beneficiary
doubt that the study would lead to their removal from the SSDI program for medical reasons.
Control Condition
Enrolled beneficiaries randomized to the control condition did not receive any study-provided
intervention services or benefits, including suspension of the medical CDR. Instead, these
participants received a resource manual that listed services and resources for persons with mental
illness available both locally and nationally. The manual, copies of which are available on SSA’s
website (http://socialsecurity.gov/disabilityresearch/mentalhealth.htm), also included the study
Chapter 2: Study Design
2-15
sites. However, in most cases, the participants were not aware of the study site’s name as
recruitment activities took place outside the study site offices. The types of services listed in each
resource manual included the following:

Public and private mental health clinics that serve individuals on a sliding scale;

Self-help and consumer support groups;

State, county, and local assistance programs that provide payment for medications for
uninsured indigent patients;

Pharmaceutical company medication assistance programs (sometimes called
“pharmaceutical programs for the indigent”);

Public health insurance options, including state Medicaid buy-in programs, the Qualified
Medicare Beneficiary program, and other federal, state, county, or local programs that
provide medical insurance to beneficiaries; and

Information on SSA work incentive provisions (such as Ticket to Work), and SSA
initiatives designed to help beneficiaries return to work.
Participants in the control group also received a modest payment to compensate for their
participation in completing interviews over their 24 months of study participation. Each participant
received a total of $100 for completion of a Baseline, seven Quarterly, and a final Followup
interview. Partial payments accompanied completion of each of the nine interviews.
Study Site Staffing
To fulfill required research tasks and fully implement the service interventions of the MHTS, each
study site employed an RA, NCC, and SE specialist. Each position provided a unique contribution
to the study to ensure that all sites achieved all research protocol and evidence-based SMM and IPS
requisites. As described in detail in Chapter 5 Implementation of Supported Employment and Other Behavioral
Health and Related Services the role of the SE specialist was to ensure the comprehensive
implementation of all aspects of IPS SE at the beneficiary level. The role of the NCC, described in
Chapter 6 Implementation of the Nurse Care Coordinator Role and Systematic Medication Management, was split
between supporting participant treatment, particularly the SMM component, coordinating medical
and psychiatric care, and completing research activities related to the treatment intervention.
Chapter 2: Study Design
2-16
The primary responsibility for the RA role was to recruit and enroll beneficiaries into the study, and
then subsequently administer the eight post-baseline interviews to all treatment and control
participants enrolled at that site. The RA was the staff anchor for both treatment and control
participants throughout their 24-month MHTS participation. Other RA tasks included compiling,
updating, and documenting all participant insurance-related data, including issues related to
participant healthcare co-pays and debit card utilization. RAs’ meticulous attention to detail is
reflected in much of the data provided in Chapter 3 Enrollment and Participation and Chapter 4 Outcomes.
Fidelity Measures
The investigators created measures of fidelity for the two key intervention components of IPS SE
and SMM. 4 Fidelity refers to the degree of adherence to the evidence-based practice. It answers the
question, “To what extent does the implementation match the model protocol?” The fidelity
measures used in the MHTS met several needs, including documentation of the nature and degree of
implementation, documentation of variations across study sites, and information needed to promote
study site improvements.
Assessment of fidelity to IPS. A designated team assessed the quality of implementation at each
study site using a 15-item measure known as the IPS Fidelity Scale (Bond, Becker, Drake, & Vogler,
1997; see Appendix 5A). The format and assessment procedures for the IPS Fidelity Scale follow the
conventions formalized in the National Implementing Evidence-Based Practice Project (McHugo et
al., 2007). Each item on this scale reflects a specific element in the practice. A 5-point behaviorally
anchored scale provides the rating range. A rating of “5” indicates close adherence to the model. A
rating of “1” represents a substantial lack of model adherence.
For quality improvement purposes, the study employed Quality Management Program Directors
(QMPD) to provide ongoing technical assistance and feedback to sites about their relative
attainment of a core element in the IPS model using the item-level fidelity ratings. In addition, the
average of the item ratings yielded a total fidelity score that expressed a global picture of overall
fidelity. The total fidelity score ranged from one (1) to five (5), with higher scores indicating more
faithful implementation.
4
The study included many other quality assurance measures and procedures for various aspects of the study, such as participant use of debit cards for
co-pays. A detailed discussion of these measures and procedures appears in Supplemental Appendix A, Study Implementation.
Chapter 2: Study Design
2-17
Assessment of fidelity to SMM. The rationale for the inclusion of SMM as part of the MHTS
treatment intervention was that it could contribute to improved outcomes by decreasing the
deleterious effects of inadequately treated illness symptoms and or medication side effects. Study site
staff received training in the principles and materials designed for TMAP, an evidenced-based
approach to medication management. Standardized assessments and documentation are used to
quantify illness symptoms, explicitly identify medication side effects, and record these observations
so as to be accessible to present and future providers. Two MHTS investigators (Drs. Bond and
Miller) developed a rating scale to measure quality of prescriber medication management of
schizophrenia in a prior project (Taylor et al., 2009). In addition, in consultation with experts in
treatment of depression and bipolar disorder, the investigators adapted TMAP scales for assessment
of quality of medication management of bipolar disorder and major depressive disorder. Medical
records served as the basis for quality assurance rating reviews. The scales were somewhat lengthy
and required the rater to search through the medical record for evidence that the prescriber had
documented intent and rationale for medication decisions and had attended to patient symptoms
and side effects. Trained RAs completed the scales on a 10 percent sample of participant records in
each quarter at each site. Westat staff randomly selected site records for review by way of an
automated process using the SMS.
Outcome Measures
The main research question addressed by the MHTS is to what extent delivering appropriate mental
health treatment and employment supports leads to better employment, health status, and quality of
life among SSDI beneficiaries with schizophrenia or an affective disorder. The primary analyses
conducted to answer this question involve a direct head-to-head test of treatment versus control
beneficiaries on the most important outcomes of interest in the study, namely employment, health,
and quality of life. However, employment was the central focus of the intervention effort, and thus it
was the focal point for measuring impact of the intervention.
Employment. The primary employment measure of interest was employment rate, specifically the
percentage of participants who reported working in a job at any point during the study period. Since
competitive employment in regular work settings is one of the seven core principles of the IPS
model of SE services, the study also measured the rate at which participants obtained competitive
jobs.
Chapter 2: Study Design
2-18
Although the primary outcome of interest to the MHTS was employment rate, there were a number
of additional characteristics of participant employment experiences of interest to the study. In
addition to analyzing outcome measures that considered all jobs a participant reported, the study
also analyzed measures that focused only on the main job 5 worked during a study period.
Definitions of these additional employment measures, analyzed for both any type of job as well as
for competitive jobs only, appear below.
5

Average weekly earnings at main job – Derived from reported weekly earnings averaged over
all reported main jobs.

Average hours per week at main job – Derived from reported hours worked per week
averaged over all reported main jobs.

Job satisfaction at main job at study exit – Derived from the level of agreement with 23
statements about personal satisfaction with a job, the work environment, and coworkers. Examples include “I feel good about this job,” “my supervisor is fair,” “my coworkers are easy to get along with,” “working conditions are good.”

Number of months to first job – Derived from all jobs reported and indicates the number of
whole months elapsed from study start (enrollment date) to the month in which a job
was first obtained.

Total months employed – Since the study collected longitudinal information directly from
participants on their employment activities since enrollment, a derived measure assessed
for each month of the study (over the 2-year period) whether participants did not work
during the month, worked part of the month, or worked for all of the month. These
monthly measures of employment status created the opportunity to construct a measure
of total months employed during the study period.

Consecutive months of employment at study exit – Derived from the endpoint of the study, this
measure computes the number of months of employment counting backwards from the
final month of study participation to the month when the participant did not report
employment. For example, if a participant did not report employment at any time
during the final month of study participation, then he received 0 months credit. If a
participant reported no employment during month 17, but did report employment
during month 18, and worked in every month until transition out of the study, then he
received 7 months credit for this variable.

Highest hourly wage – This measure is the highest hourly wage reported across all jobs held
during the study.
The main job was defined as the job at which a participant worked the longest or worked the most hours during the reporting period for a respective
interview. For most participants, interviews were spaced approximately every three months, however, if the previous interview was skipped,
questionnaire items related to employment outcomes were asked since the date of the last completed interview (i.e., past six months).
Chapter 2: Study Design
2-19
Health status. While participation in employment was the primary outcome of interest for the
study, SSA was also interested in knowing whether access to the intervention would result in
improved health status. The SF-12 provided the measures of health status for the MHTS.
The SF-12 is a well-known self-report health survey empirically derived from the longer SF-36.
Designed for use in clinical research, health policy evaluations, and general population surveys, the
SF-36 is the “gold standard” in self-reported health assessment (Ware, Kosinski, & Keller, 1996).
The SF-12 serves to reduce respondent burden while maintaining acceptable precision. The eight
health concepts measured by the SF-12 are:
1.
Limitations in physical activities due to a health problem;
2.
Limitations in social activities due to a health problem;
3.
Limitations in usual role activities due to a physical health problem;
4.
Limitations in usual role activities due to an emotional problem;
5.
Pain;
6.
General mental health;
7.
Vitality; and
8.
General health perceptions.
The SF-12 yields two summary measures—a Mental Component Score (MCS) and a Physical
Component Score (PCS). Both the MCS and PCS were assessed at baseline and again at study exit
to assess improvements in health and functioning over time, as well as determine if treatment group
participants showed more improvement than control group participants. Refer to Appendix 2A for
a copy of the SF-12 items and instructions on how to derive the MCS and PCS summary scores
(Ware et al., 2002).
Quality of life. Previous research suggests that individuals who maintain competitive employment
over time report higher quality of life (Bond et al., 2001; Mueser et al., 1997). The MHTS measured
quality of life with a single scale item from the Modified Lehman Quality of Life Inventory (QOLIM). The QOLI-M is a shortened version of the well-known QOLI (Lehman, 1988), and has been
used in four IPS studies (Bond, 2007; Drake et al., 1999; Drake, McHugo et al., 1996; Mueser et al.,
2004). Participants were asked to rate how they felt about their life in general using a 7-point scale:
(1) terrible, (2) unhappy, (3) mostly dissatisfied, (4) mixed, (5) mostly satisfied, (6) pleased, and
Chapter 2: Study Design
2-20
(7) delighted. Similar to the health status measures described above, the quality of life measure was
assessed at baseline and study exit.
Secondary outcomes of interest. The primary outcomes of interest to the MHTS to determine the
effectiveness of the intervention include employment rate, mental and physical health status, and
quality of life. However, the scope of the planned outcomes analyses expanded to include additional
secondary outcomes including employment-related measures (e.g., hours worked per week, wages);
earnings and income measures; and healthcare service utilization measures.
Impacts of the MHTS intervention on health and mental health service utilization levels and
patterns are potentially important to SSA from a budgetary and social cost standpoint. There is a
substantially reduced net cost to government if the intervention reduces overall costs of these
services. In order to estimate the intervention impacts on service use in the MHTS, secondary
outcome measures were constructed that compared participant reports of service utilization for
several different kinds of healthcare services during the two-year study period. The measures of
service utilization studied include number of emergency room visits, number of hospital admissions,
number of nights spent in the hospital, number of visits to receive outpatient psychiatric
emergency/crisis services, and number of visits to other mental health providers (e.g., clinics,
therapists, etc.).
Data Collection Procedures
The principal source of data used to test the study hypotheses were self-reported measures collected
directly from enrolled beneficiaries in the treatment and control groups through computer-assisted
personal interviews (CAPI). Beneficiaries participated in nine interviews evenly distributed over the
2-year period of study participation. The initial Baseline interview occurred at enrollment, before
randomization to a study condition. All Baseline interviews were conducted in person. Each
participant subsequently completed seven Quarterly interviews covering the interim months of study
participation. These interviews were shorter than the Baseline interview and collected only the most
pertinent information needed to assess the study outcomes. These interviews were either in person
or over the telephone, depending upon the circumstances of the particular participant. At the end of
the two-year study period, participants participated in a final Followup interview. Conducted either
over the telephone or in person, this interview mirrored the more extensive Baseline interview.
Chapter 2: Study Design
2-21
The Baseline interview included questions about the participant’s demographic information, work
history, benefit status, attitudes toward work, health care coverage, health care utilization, and
prescription medication use. A final section of the Baseline interview included administration of a
paper-pencil version of the Digit Symbol Test. The test is a reliable and efficient measure of
information processing efficiency (Dickinson, Ramsey, and Gold, 2007). It served as a brief measure
of enrolled beneficiaries’ cognitive functioning.
The seven Quarterly interviews (conducted every three months) included questions about each
participant’s contact information, demographics, health status, alcohol and substance use,
employment outcomes and income, job seeking behavior, health care service utilization, and quality
of life. Quarterly interviews were not the same across the treatment and control groups. In order to
reduce respondent burden and minimize non-response, the seven Quarterly interviews for
participants in the control group included only the same battery of items covering employment and
healthcare service utilization. Additional questions asked of treatment group participants included
questions related to health and functioning, and alcohol and drug abuse.
The final Followup interview collected once again particular demographic information about the
participant, and information about employment outcomes and income, health status, alcohol and
substance abuse, attitudes toward work and SSA benefits programs, health care service utilization,
and quality of life. Interviews with the treatment group also included questions about participant
satisfaction with the services they received during the study. The Baseline and final Followup
interviews were essentially the same with the exceptions that the Baseline interview collected
information on work history prior to study enrollment, and the final Followup interview collected
information on employment outcomes since the last completed interview. Table 2-2 provides an
overview of the content and timing of these self-report interviews.
The Supplemental Appendix includes a copy of the questionnaires that were developed for the
Baseline, Quarterly, and final Followup interviews.
Chapter 2: Study Design
2-22
Table 2-2.
Content domains for the Baseline, Quarterly, and final Followup interviews
Measures
Demographic Information
Baseline
Quarterly1
Followup





Work History

Employment Outcomes
Income Review



Health Status (SF-12)



Quality of Life (Item from Lehman QoL Inventory)



Alcohol and Drug Use (Addiction Severity Index)



Health Care Coverage

Health Care Service Utilization

Attitudes Toward Work

1
2
2



Control condition participants received only the questions about Employment Outcomes and Health Care Service Utilization in the
Quarterly interviews.
Only control condition participants received questions about Health Care Coverage.
Interview Completion Rates
Table 2-3 shows the interview completion rates by month and study group. The overall interview
completion rate for the followup interviews was 84 percent. The overall completion rate for the
treatment group was 81.6 percent, ranging from a low of 78 percent in Quarter 7 to a high of 86
percent in Quarter 1. The overall completion rate for the control group was 86 percent, with a range
of 84 percent to 89 percent. There were fewer completions by the treatment group than by the
control group at every quarter after baseline. The difference in attrition rates between arms grew
steadily from Quarter 1 to followup and was statistically significant for every quarter after Quarter 5.
As discussed previously in this chapter, participants in the control group were paid a nominal fee for
the Baseline interview (after being assigned to the control group) and small payments per interview
beyond the baseline. Due to the rich nature of the intervention, treatment group participants did not
receive payments for any interviews.
Chapter 2: Study Design
2-23
Table 2-3.
Interview completion rates by quarter and study group
Interview
Baseline
Treatment
(N=1,121)
freq
%
1,121
100.0
Control
(N=1,117)
freq
%
1,117
100.0
Quarter #1
Quarter #2
Quarter #3
Quarter #4
Quarter #5
Quarter #6
Quarter #7
Followup
Total**
962
942
920
918
909
890
876
902*
7,319
988
964
956
947
956
942
933
991*
7,677
85.8
84.0
82.1
81.9
81.1
79.4
78.1
80.5
81.6
88.5
86.3
85.6
84.8
85.6
84.3
83.5
88.7
85.9
Total
(N=2,238)
freq
%
2,238
100.0
1,950
1,906
1,876
1,865
1,865
1,832
1,809
1,893
14,996
87.1
85.2
83.8
83.3
83.3
81.9
80.8
84.6
83.8
p-value
0.0627
0.1308
0.0239
0.0667
0.0043
0.0024
0.0012
0.0000
The figures on completion rates include five interviews (2 treatment group and 3 control group) found to be unusable due to a laptop
computer malfunction.
** Excludes baseline.
*
Data Preparation and Adjustments
As is typical of large controlled trials, events occur that result in participants failing to complete all
phases of a study. Some participants drop out of the study, others simply do not return for various
activities, and some participate only sporadically. Beneficiaries in the MHTS were no different. The
study experienced participants who withdrew for a variety of reasons (including they left the
country, became disillusioned or lost interest, etc.) or participants who died. The study dropped
some in the treatment group because they failed their GME or refused to complete it. All of these
situations give rise to concerns about attrition bias and reduced precision and power.
The Baseline interviews and every Quarterly interview prior to attrition provide substantial
information about the types of people who stay in the study or left the study before the end of 24
months. The follow-along interview patterns also provide information about those beneficiaries who
participated only sporadically. Of course, it is unknown what events following the last interview of a
beneficiary who stopped participating led him or her to the decision to stop participating. There is
concern that participants who dropped out of the treatment group were less motivated to find
employment than those who remained in the study. If true, comparisons between treatment and
control groups may be exaggerated relative to an ITT analysis without attrition.
Two data adjustment methods addressed this issue. First, participants who did not complete at least
two post-baseline interviews were “non-respondents” and these participants were not included in
Chapter 2: Study Design
2-24
the analysis. Instead, weights were created to adjust for their removal. Second, imputations provided
a solution for other participants with substantial missing data. Weighting adjustments for severe
non-response and imputation for mild non-response were employed for specific reasons. Weighting
is better than imputation at avoiding biases in complex multivariate relationships, but it makes very
light use of the collected data on cases that are classified as non-respondents. Imputation makes very
strong use of partial data, and therefore yields better estimates of marginal means and differences
between pre-identified subgroups than can be obtained through weighting. Thus, the rationale for
weighting was that the limited participation of non-respondents resulted in insufficient partial data
to justify imputation-induced biases in complex multivariate analyses. The rationale for imputation
was so much partial data existed with the respondents that using weighting would result in
unacceptable variance increases. Below is a more detailed description of the methods for these two
procedures.
Weighting Adjustments
As indicated in the preceding section, non-respondents were participants who did not complete at
least two interviews beyond the Baseline interview. There were 154 “non-respondents” in the study
with less than two post-baseline interviews. (These 154 non-respondents include those treatment
cases that were dropped from the study on account of a failed or incomplete GME.) However, the
non-respondents group also included another five participants who completed two or more
interviews, but whose data were corrupted due to a malfunctioning interviewer laptop computer.
The weights for these 159 participants were set to zero and the remainder of the study sample
(N=2,079) was weighted to reflect these individuals. An additional 24 participants (11 treatment
group participants and 13 control group participants) who died during the study completed more
than two interviews after baseline. Although not considered to be “non-respondents,” these
individuals were excluded from the analyses included in this report. Thus, there were a total of 2,055
participants included in the analytic sample used for this report.
The weighting process involved two steps. First, demographic characteristics and baseline measures
potentially associated with likelihood of study withdrawal defined 20 subgroups, across which studywithdrawal rates widely varied. Five of these 20 had perfect response rates. Response rates in the
other 15 dipped as low as 64 percent. The 20 subgroups also varied in size from 31 to 447
participants. Separate subgroups were defined for the treatment and control groups, so as to avoid
Chapter 2: Study Design
2-25
transfers of weight across the two groups. Some of the baseline variables related to attrition
included:

Receipt of Temporary Assistance for Needy Families,

Desired number of work hours per week,

Desired pay,

Perceived quality of life,

Number of children in the household,

Number of adults in the household,

Hospital utilization,

Outpatient mental health utilization,

Use of psychotropic drugs,

Current work status (at baseline), and

Severity of substance abuse.
Second, calculations with the following formula provided non-response adjustment factors within
each subgroup:
Adjustment factor =
Number of enrolled participants
Number of responding participants
All enrolled beneficiaries received an initial weight of 1.0. The adjustment factors were applied to the
initial weights of respondents to increase their weight upward for the non-responding beneficiaries.
Imputation
There are two key reasons why imputing missing values is critical in this study. Most statistical
procedures drop entire cases where any of the variables required for a particular analysis contain
missing values. For simple marginal means, two-way tables, and bivariate correlations, this is
sometimes an acceptable practice, but for analyses that draw on many variables simultaneously, a
handful of cases with one variable missing and another handful of cases with a different variable
missing soon add up to large numbers of dropped cases. For the MHTS, many of the outcome
Chapter 2: Study Design
2-26
variables require data across all the waves of data collection, and so drop-case rules would result in
the loss of a large quantity of data. This loss of data would have resulted in reduced precision and
power, and potential attrition bias. Imputation helps to correct both of these problems. Many
records deleted because of missing data actually have substantial partial data that can be used to
impute the values that are missing. Using this partial information helps to improve power and
precision.
The issue of attrition bias is particularly important in studies with an ITT model (such as the
MHTS) with the expectation that the analysis accounts for all cases. While it is generally impossible
to understand the full details of the missing data mechanism, the study did gather some valuable
clues. As is discussed in the final section (Imputation Procedures) of Appendix 2B, subjects who
missed one or more waves but participated in the final Followup interview had many more
hospitalizations, and subjects who dropped out of the survey for good at some point after baseline
had poorer job-seeking results than those who persisted in the survey. Since the intervention tends
to reduce hospitalization and increase the probability of finding a job, a simple complete-case
analysis would probably exaggerate treatment effects. A well-done imputation procedure can reduce
the risk of such biases due to informative non-response. Weighting can also reduce the risk of bias,
but discards much more partial data and is thus far less efficient.
Imputation inflates some differences and deflates others. There is no way to determine whether the
differences based on imputed values are better or worse for any particular analysis than those
obtained with only reported data. They will certainly be different. At the same time, it is most likely
that variances on estimated differences will be smaller with imputation than without. For some
analyses, the reduction in variance will be substantial, while for others it will be modest.
There is a tendency to weaken associations between variables with imputation. This would suggest
that it is more likely for imputation in this case to have deflated treatment effect estimates than to
have inflated them. Even where this happens, the reduction in variance may be enough to offset the
depression of the estimated effect so that one still obtains a more significant test statistic for the
treatment effect.
The methodology used in the imputations is described in Krenzke and Judkins (2008). The
methodology combined traditional methods for hot deck imputation with modern model-dependent
chained parametric procedures. Several research papers have shown performance better to the best
alternative software systems available. The post-baseline interviews (of which there are eight—seven
Quarterly and a final Followup) are the source of all outcome data for the study. The most
Chapter 2: Study Design
2-27
important imputations are for those participants who did not complete a final Followup interview
but did complete the Baseline interview plus at least two post-baseline interviews. For these
participants, imputations included all of the information that should have been collected after the
last interview that they completed. In addition, there were imputations for a small number of
missing values for a limited set of items for participants who missed one or more quarters but who
returned to complete the final Followup interview. For these participants, the imputation included
income components for the missed quarters, attributes of the main jobs held in the missing quarters
(provided that the catch-up interviews 6 reported employment during the period), and (for the
treatment group only) updated demographics and substance abuse information.
Fifty-eight percent of the study sample required no imputation at all. Nine percent required imputing
the final Followup round and possibly additional quarterly rounds. Thirty-three percent required
imputation of only scattered items not subject to extended recall. These figures are typical for a
complex study with many moving pieces in the data collection system. Marker, Judkins, and Winglee
(2001) report on the imputation of costs and payments for medical events in the Medicare Current
Beneficiary Survey. In that survey, 96 percent of persons in the best-reporting domain of interest
had to have at least one cost or payment amount for at least one medical event over the course of a
year imputed. Without the imputation, it would have been impossible to assemble reasonable
statistics about person-level costs and payments.
Appendix 2A contains a complete description of the imputation procedure and more information
on the likely impact of the imputation for a few analyses. While some analyses were conducted
separately for cases with nearly complete records and cases requiring major imputation, not much
time was devoted to this effort. The results definitely are sensitive to the imputation procedure, but
given that non-respondents appeared to be sicker and less successful in their job hunts than were
respondents, there is little reason to pay attention to analyses based only on good respondents.
Statistical Procedures
Data analyses presented in this report use the weights described the previous section, unless
otherwise specified. Statistical Analysis Software (SAS) procedures for analyzing data were used to
analyze the weighted data, notably SURVEYFREQ, SURVEYREG, and SURVEYLOGISTIC.
6
During each interview, participants were asked if they worked since the date of the last completed interview allowing for the collection of complete
data on employment outcomes even if interviews were skipped. For example, if they had skipped the previous interview (three months ago), they
were asked about work in the past six months as opposed to the normal three months.
Chapter 2: Study Design
2-28
These procedures adapt standard statistical procedures to incorporate unequal weighting
adjustments for non-response.
Chi-square tests served to test for differences between treatment and control groups for categorical
data. In cases where the outcome measure resulted in continuous data, an analogue of the Wilcoxon
test served as the test for differences between treatment and control groups. Specifically, data were
ranked in the combined study groups then weighted t-tests (computed in SURVEYREG) were used
to test for differences between ranks in treatment and control groups. Since most measures
considered in many analyses were non-negative and some had clear outliers, normality was not a
reasonable assumption. The Wilcoxon test is a well-known solution to overcome the need to assume
normality in the distribution of outcome scores (Conover and Iman, 1981; Zimmerman, 1992).
The study had the potential for exploring outcomes in substantial depth, given the rich nature of the
data collected over the 2-year study period. Among the many classes of variables that could account
for particular outcomes were clinical variables (e.g., diagnosis or cognitive functioning), demographic
variables (e.g., age and education), baseline characteristics of participants (such as health status and
recent work history), among others. In an attempt to provide a better understanding of the
outcomes, the study required use of multivariate techniques to explore relations among large
numbers of potential predictor variables. Several multivariate analyses used multiple logistic
regression technique with binary dependent variables, such as obtained employment (yes vs. no), to
explore and improve understanding of the outcomes.
The study employed the zero-inflated negative binomial regression technique to analyze the
outcome number-of-months employed (in Chapter 4) and the count of healthcare service utilization
variables (in Chapter 8). The rationale for using this estimation method was based primarily on the
distributional characteristics of the dependent variables, in particular, the facts that they are all count
(integer) data and that they are characterized by a large mass of observations at zero. The zeroinflated negative binomial regression model assumes that the data are a mixture of two separate data
generation processes—one generates only zeros and the other is a negative binomial data-generating
process. For the quality of life outcome measure, an ordered logistic regression model was run. This
model estimates categorical outcomes and expresses probabilities of each outcome as subtraction of
cumulative probability.
All bivariable models included a forced variable (called the “treatment dummy”), indicating whether
individual cases referred to a treatment intervention case or a control case, along with each of 15
independent (predictor) variables. Candidate variables for the multivariable model were those that
Chapter 2: Study Design
2-29
showed a marginal association with the dependent variable at the 0.25 level of significance after
adjustment for the treatment dummy. The multivariable, multiple logistic regression procedures
included both backwards and stepwise regression to ensure that disparate models did not occur.
Discussion of final models focused on any independent variables showing a 0.05 level of statistical
significance.
In an attempt to provide an even better understanding of the outcomes, investigators also used
multivariate techniques to explore relations among an additional 13 potential predictor variables that
were applicable for treatment group participants only. Additional predictor variables included for
treatment group participants only included dummy variables indicating use of some other behavioral
health or related service (e.g., substance abuse treatment, case management, social skills training,
etc.) at some point during the 24-month study period; variables measuring the extent to which a
participants was engaged with the IPS and SMM components of the intervention; size of study site
(measured by the number of clients served at the site); and the site fidelity scores.
Data Limitations
The study results rely almost exclusively on the self-report data collected in the eight post-baseline
interviews. The study RAs, NCCs, and the QMPDs provided additional data on participants in the
treatment group that served a number of treatment group only analyses, and some analyses relied
upon administrative data provided by SSA. However, the self-report data collected in interviews
with the treatment and control groups served to answer the primary research question concerning
the impact of the intervention on employment, health, and quality of life. Data provided from all of
these sources have particular limitations that require acknowledgement.
The study results requiring use of SSA administrative data presented in this report are limited to the
accuracy of that data. SSA’s administrative data did in fact contain errors of accuracy, much of it due
to the currency of the data. For example, during and after the enrollment process, a few beneficiaries
revealed that they had legal guardians, thus making them ineligible for the study. Yet the
administrative data did not indicate this fact, more than likely because the data were at best several
months old when recruitment efforts began. This example serves to remind the reader that the
administrative data contain some inaccuracies.
Chapter 2: Study Design
2-30
The study development process resulted in the production of 14 manuals (e.g., recruitment and
enrollment, conducting RIG meetings, documenting MHTS insurance coverage) explicating precise
procedures for documenting and carrying out study activities. In addition to the manuals, the study
provided training, ongoing discussion groups, and periodic monitoring to check on and improve
reliability. However, despite these extensive efforts and precautions, RA, NCC, and QMPD errors in
accurate recording or omission inevitably occurred.
Limitations associated with the interview data are likely the greatest concern to SSA and others
interested in the results of the MHTS. Construction of the CAPI instrument used for the self-report
interviews followed standard procedures for developing content, translating the interview content
into programming specifications (including appropriate skip patterns), testing the CAPI, and
deploying the instrument into the field. Each study site received a laptop computer, assigned to its
RA(s), preloaded with the interview and customized with a limited amount of beneficiary
information for individuals residing in the respective study site catchment areas (study ID and
name). A CAPI interviewer manual and a two-day training session provided RAs with the
preparation and practice necessary to conduct the interviews. However, because of a 5-month delay
in the study start, RAs received refresher training and practice closer to the start of recruitment. RAs
also participated in telephone conference calls to receive ongoing support and technical assistance
related to administering and completing the CAPI interviews. RAs transmitted interview data from
their CAPI laptops to Westat’s home office daily.
Chapter 2: Study Design
2-31
Chapter 3
Enrollment and Participation
The intervention package provided to treatment group participants in the Mental Health Treatment
Study (MHTS) combined a number of interventions, three of which included (1) the Individual
Placement and Support (IPS) model of supported employment (SE); (2) systematic medication
management (SMM); and (3) enhanced insurance coverage for behavioral health services. Based on
prior evidence with these interventions, one might expect them to yield positive employment and
health outcomes for Social Security Disability Insurance (SSDI) beneficiaries with schizophrenia or
an affective disorder. However, no tests exist with these interventions, either individually or in
combination, on a target population primarily comprised of this group of beneficiaries. Thus, the
focus of the MHTS was to carry out such a test to determine the effectiveness of this intervention
package and thus the likely consequences of a Social Security Administration (SSA) policy initiative
to make this package of interventions available on a national basis.
Of course, to understand the consequences of making the MHTS interventions available, the initial
question that needed addressing was, if offered, who will take advantage of these services? This
question was critical to understanding the generalizability of findings from the MHTS. It was also
critical to understanding the policy implications for SSA and other agencies serving individuals with
severe mental impairments, and it emphasized the concern that SSA had for providing SSDI
beneficiaries with access to quality mental health services.
This chapter addresses the extent of participation in the MHTS from the SSDI rolls. It is worth
noting that not everyone eligible to participate in the MHTS agreed to enroll in the study, so it is not
appropriate to draw inferences about all SSDI beneficiaries with schizophrenia or an affective
disorder from the results presented in this report. However, the characteristics of those SSDI
beneficiaries that did agree to participate and those who did not needs to be studied to help design
strategies to increase participation in the event that the SSA or another agency intends to implement
the treatment intervention on a wider scale. This section presents extensive comparisons between
the characteristics of beneficiaries who enrolled in the MHTS to those of beneficiaries who also had
the opportunity but chose not to enroll. These analyses provide SSA with an understanding of the
universe to which one might appropriately generalize the findings of the MHTS. In addition, this
understanding provides clues as to who might agree to participate in a broader implementation of
the intervention.
Chapter 3: Enrollment and Participation
3-1
A second set of analyses characterizes the beneficiaries who enrolled in the study, and examines the
similarities and differences between participants assigned to each arm of the study. A fundamental
expectation of the random assignment process was equivalence between the treatment and control
groups on variables important to the investigation. Using data collected at baseline (which was prior
to randomization), an assessment was made of the success in creating two equal groups of
participants. This assessment included a comparison on demographic variables, SSDI program
variables, and key measures of employment, health status, and income.
Are There Differences Between SSDI Beneficiaries Who Enrolled and Those
Who Did Not Enroll in the MHTS?
Since the MHTS was a randomized controlled trial, eligible beneficiaries only had a 50 percent
chance of receiving the intervention package, whereas an ongoing program of services would not
have this feature of a randomized assignment. Even when interested in participating, some eligible
beneficiaries may have declined knowing they only had a 50 percent chance of receiving the
intervention services offered to the treatment group.
The approach to answering this important policy question consisted of the following steps:
1.
Assign the 61,530 beneficiaries residing in the catchment areas of the 23 study sites to
potential enrollment groups based on knowledge gleaned from the recruitment effort.
2.
Prepare a test and validation sample.
3.
Estimate predictors of enrollment using logistic functional form and maximum
likelihood estimation on the test and validation samples.
Assignment of Beneficiaries to Enrollment Groups
As described in Chapter 2, the recruitment process involved several attempts by the study Research
Assistant (RA) in each study site to contact beneficiaries in their catchment area. First, a letter was
sent to the beneficiary’s address of record. Following the letter, one or more telephone calls were
made to make verbal contact with the beneficiary. Interested beneficiaries were asked to attend a
Research Information Group (RIG) meeting in person to obtain further information about the
study. A beneficiary could refuse the study at any of the three steps. However, to join the study, a
beneficiary was required to attend at least two RIG meetings.
Chapter 3: Enrollment and Participation
3-2
The RA at the sites following release of beneficiaries’ names and contact information documented
all key recruitment activities in the web-based Study Management System (SMS). These data served
as the basis for classifying beneficiaries into one of three potential enrollment groups—potential
enrollee, possibly potential enrollee, or not a potential enrollee. Figure 3-1 presents the logic model used to
assign beneficiaries across all 23 study sites to a potential enrollment group. Using the information
recorded in the SMS that documented the RAs recruitment efforts, MHTS investigators were able to
follow a beneficiary’s path throughout the recruitment process. The total eligible population
identified from SSA administrative records as meeting the inclusion criteria and available to the
study sites for recruitment into the study was 61,530. Of the 61,530 SSDI beneficiaries that were
included in the list sample received from SSA, RAs spoke with only 17,642 beneficiaries. Of those
17,642 beneficiaries contacted via telephone to determine their interest in attending a RIG meeting
to learn more about study participation, only 3,971 agreed to attend a RIG meeting, with 2,238
beneficiaries subsequently enrolling into the MHTS after attending at least two RIG meetings. The
figure shows the final disposition of those who did not attend a RIG meeting, as well as the final
disposition for those who attended a RIG meeting but did not enroll into the study.
Figure 3-1 also presents the final disposition at the end of all recruitment efforts for beneficiaries
who never had an opportunity to hear about the MHTS, for a variety of reasons. In some cases, RAs
screened out beneficiaries who were found to be ineligible before recruitment efforts began
(N=3,614). In other cases, RAs were never able to reach beneficiaries over the telephone to invite
them to a RIG meeting, in spite of various call attempts (i.e., labeled as “max call attempts”). RAs
were not able to locate almost one-quarter of beneficiaries included in the sample (N=14,397).
Although SSDI beneficiaries receive monthly payments from SSA, the majority of beneficiaries
receive their cash benefit in some electronic form (e.g., direct deposit), as opposed to receiving an
actual check in the mail; so in many cases, beneficiary mailing addresses were out-of-date. Toward
the end of the recruitment period, some beneficiaries were not contacted via telephone even though
an introductory letter was mailed to them (i.e., labeled as “letter sent, no followup”) because the
site’s recruitment target had been met. Some RAs preferred to use hard-copy materials to manage
their recruitment tasks, and then subsequently document the results in the SMS. However, not all
RAs were diligent about documenting their efforts in the SMS; at the end of the recruitment period
there was no electronic call record in the SMS for 10,171 beneficiaries.
Chapter 3: Enrollment and Participation
3-3
Figure 3-1.
Logic chart for assigning beneficiaries to enrollee group classifications
All beneficiaries (N=61,530)
Letter mailed (N=57,364)
No letter mailed (N=4,166)
Did not speak to beneficiary (N=39,722)
Spoke to beneficiary (N=17,642)
Attended RIG
meeting (N=3,971)
Did not attend RIG
meeting (N=13,671)
Ineligible
(N=1,499)
Enrolled
Refused
Ineligible
Other2
(N=2,238)
(N=684)
(N=161)
(N=892)
Potential Enrollees (N=15,982)
Test (N=7,991)
Validation (N=7,991)
No call
record
(N=10,171)
Max call
attempts
(N=3,592)
Max call
attempts
(N=4,022)
Ineligible1
(N=3,614)
Refused
No followup
(N=7,419)
(N=728)
Possibly Potential Enrollees (N=13,763)
Test (N=6,882)
Letter sent, no
followup
(N=7,948)
Could not
locate
(N=14,397)
Not Potential Enrollees (N=31,785)
Validation (N=6,881)
1
These beneficiaries were found to be ineligible in a variety of ways, including reviews of study site records revealed the
beneficiary was receiving SE services at the site; a proxy answered the recruitment call and indicated the beneficiary has a
legal guardian, etc.
2
The “Other” group who attended a RIG meeting includes beneficiaries who fall into one of three result codes: No longer
locatable (N=74), Max calls (N=731), and Unknown (N=87).
Using the information recorded in the SMS indicating the final disposition of each recruitment case,
MHTS investigators were able to classify all 61,530 beneficiaries included in the original sample that
was received from SSA into one of three potential enrollment groups. The criteria for classifying a
beneficiary as a not potential enrollee were (1) absence of a contact date (most likely because release of
the beneficiary information never occurred) or (2) presence of a result code that indicated that the
beneficiary was “ineligible,” “activated, not recruited,” or “not locatable.” The criteria for classifying
the beneficiary as a potential enrollee were (1) presence of a contact date or (2) the result of a contact
attempt coded as either “completed” or “refused,” or (3) there was a recorded “spoke to beneficiary
date.” The criteria for classifying a beneficiary as a possibly potential enrollee were (1) presence of a
contact date and (2) recorded result was “unknown” or “max calls.”
Beneficiaries contacted and spoken to by study site staff comprised the 15,982 potential enrollees.
These individuals had at least one opportunity to hear about the study, even though they may or
may not have attended a RIG meeting. However, nearly a quarter of them (24.8%) attended a RIG
Chapter 3: Enrollment and Participation
3-4
meeting before making a decision to enroll in the study. The 31,785 not potential enrollees included
beneficiaries who had no opportunity to participate in the study or (presumably) even know it
existed. Those beneficiaries considered to be out of the pool of potential enrollees included
individuals for whom the SMS indicated they were “ineligible” (beneficiary was deceased, moved out
of the catchment area, was already receiving services at the study site in their area, etc.), or not
recruited because the study site “could not locate” the beneficiary.
The 13,763 possibly potential enrollees were a unique group of beneficiaries for whom no concrete
information about their interest in the study existed. The largest beneficiary group in this category
(10,171) had a final enrollment result code of “no call record,” suggesting that there was no
indication of a date on which a recruiter actually spoke to the beneficiary. The other group
comprising the possibly potential enrollees included 3,592 beneficiaries with a final enrollment result
code of “max call attempts” (again with no indication of a “spoke to beneficiary” date). In these
cases, the RA was able to identify a phone number that applied to the beneficiary, and in some of
these cases, recruiters indicated that a telephone message was left, but there was no indication that
anybody spoke to the person or that the beneficiary ever received the message. However, it was not
technically correct to classify the beneficiary as ineligible or not locatable. Thus, it was appropriate to
create a third category, namely possibly potential enrollee.
Preparation of Test and Validation Samples
The goal of the enrollment analysis was to identify predictors that could convey what characteristics
of beneficiaries were indicative of study enrollment. To achieve this end, the analysis included only
two of the three groups of beneficiaries, those classified as potential enrollees, and those classified as
possibly potential enrollees. Potential enrollees were those beneficiaries for whom there was clear evidence
that they had received some personal contact about joining the study and either enrolled or did not.
At least they knew that the study existed and that they were eligible for it. Possibly potential enrollees
included beneficiaries for whom such clear evidence of personal contact was lacking, but for whom
clear evidence of not receiving a personal contact was also lacking. Of course, none of the
beneficiaries in this second group in fact enrolled in the study. However, the investigative team
could not ensure that these beneficiaries never received an opportunity to know about the study.
The not potential enrollee group was set aside for this analysis because the vast majority of them had no
opportunity to know about the study. The others were not eligible.
Chapter 3: Enrollment and Participation
3-5
The groups of potential enrollees and possibly potential enrollees were randomly split into two roughly equal
halves. One-half became a test sample, and the other a validation sample (See Figure 3-1). The
subsequent analyses used the two test samples to explore predictors of enrollment. The validation
sample corroborated the results. Recognizing that the results between the two sets of data would
likely vary, only those predictors with a significant relationship to the dependent variable (actual
enrollment into the study) were retained in the validation sample results.
Estimation of Regression Models Predicting Enrollment
Regression analysis was used to identify the factors that predicted who enrolled in the study (and
who did not). Exploratory logistic regressions were estimated with the test samples (of potential
enrollees and of potential plus possibly potential enrollees) using a variety of estimation methods, including
maximum likelihood with separate site intercepts, conditional (fixed-effects) maximum likelihood,
maximum-likelihood with clustering of errors (by site), and random-effects logistic regressions (with
random intercepts for each site). A variety of explanatory variables were included in the test
regressions, including (but not limited to) those shown in Table 3-1. (See Appendix 3A for a list and
description of all variables included in the regression models. See Appendix 3B for descriptive
statistics from the regression analysis cases.) A number of different variables were included in
preliminary versions of the exploratory regressions on the test sample but were dropped from the
final exploratory regressions, including the dummy variables for year of recruitment, the variables
pertaining to serious mental illness diagnosis from SSA’s Master Beneficiary Record (MBR), and
county unemployment rates. Other variables tested, but not included in the final exploratory
regressions (reported below), included additional proxy socioeconomic variables from (1) the Census
2000 SF3 files and the American Communities Survey files, (2) alternative definitions for Disability
Control File earnings variables, and (3) alternative functional forms for variables such as age and
distance from the study site. Note also that the participation regressions did not include any of the
variables from the Baseline interview since these were only available on beneficiaries who did in fact
decide to enroll in the study.
Chapter 3: Enrollment and Participation
3-6
Table 3-1.
Results of logistic regressions on MHTS participation (Enrolled = 1)1
Test sample
Validation sample
Potential plus
possibly
potential
enrollees
(n=14,513)
p-value
Potential
enrollees
(n=7,933)
Coeff p-value
Potential plus
possibly
potential
enrollees
(n=14,637)
Coeff p-value
-0.002 <0.001
-0.002 <0.001
-0.002 <0.001
-0.002 <0.001
0.276
0.026
0.376
0.002
-0.039
0.749
0.006
0.957
-0.103
0.009
-0.126 <0.001
0.004
0.922
-0.002
0.960
Gender (Male)
0.221
0.001
0.124
0.060
0.135
0.054
0.091
0.176
SSI
-0.194
0.033
-0.182
0.037
-0.194
0.039
-0.216
0.017
Predictor variable
Months on Rolls
Age (days) x
(Age)2
x
10-3
10-7
Rep Payee
Potential
enrollees
(n=7,815)
Coeff p-value
Coeff
-0.487 <0.001
-0.760 <0.001
-0.573 <0.001
-0.806 <0.001
0.038 <0.001
0.053 <0.001
0.007
0.457
0.027
0.004
-0.045
0.002
-0.055 <0.001
0.003
0.860
-0.014
0.327
Some College (census tract)
0.542
0.023
0.447
0.049
-0.130
0.597
-0.108
0.648
Median Income (census
tract)
-0.374
0.005
-0.285
0.025
-0.062
0.651
0.023
0.863
Race (Black)
0.286
0.001
0.223
0.007
0.274
0.002
0.292 <0.001
Ln of Distance from Site
-0.114
0.005
-0.141 <0.001
-0.111
0.008
-0.119
0.003
Had Active Ticket (w/in 90
days)
0.290
0.193
0.216
0.135
0.536
0.118
0.575
Had Active Ticket (ever)
0.839 <0.001
0.857 <0.001
0.725 <0.001
0.841 <0.001
Trial Work End Date (10
yrs.+ ago)
0.398
0.004
0.367
0.006
0.523 <0.001
0.517 <0.001
Trial Work End Date (5-10
yrs. ago)
0.336
0.014
0.296
0.024
0.139
0.352
0.130
0.367
Trial Work End Date (0-5 yrs.
ago)
0.309
0.062
0.334
0.029
0.325
0.072
0.229
0.181
Trial Work End Date (0-3 yrs.
post-recruitment date)
1.183 <0.001
1.081 <0.001
0.726
0.013
0.839
0.002
Sq. Root of Reported
Earnings (1-6 mos. prerecruitment)
-0.014 <0.001
-0.019 <0.001
-0.020 <0.001
-0.022 <0.001
Sq. Root of Reported
Earnings (7-23 mos.
pre-recruitment) x 10-1
0.038
0.014
0.029
0.036
0.013
0.002
No Earnings Report (1-6
mos. pre-recruitment)
-0.657
0.002
-0.653
0.001
-0.914 <0.001
-0.931 <0.001
Constant
0.129
0.937
-2.443
0.112
0.040
-2.246
Time to Recruit x
(Time to
10-1
Recruit)2
x
10-4
1 The
0.294
0.435
0.981
0.922
0.158
reduced Ns are a result of the regression analysis excluding beneficiaries with missing or undefined values for variables included in
the regressions, or with negative values that were clearly erroneous. Such exclusions constituted less than 10 percent of all cases in
each of the groups shown above.
Chapter 3: Enrollment and Participation
3-7
Variables were retained in the exploratory regressions based on their p-values and (in select
cases) the consistency of the regression results with prior expectations, evidence, or theory. Since
results for individual explanatory variables were not always very similar between the potential enrollee
exploratory regressions for the test sample and those for the potential enrollees plus possibly potential
enrollees, the general approach was to retain variables that met the criteria in at least one of the two
regressions for testing in the validation sample.
Table 3-1 shows the final versions of the exploratory regressions on the test sample. Reported
regression results include the coefficients for each explanatory variable. 1 All of the explanatory
variables except one have two-tailed p-values < 0.1 for both potential enrollees and potential enrollees plus
possibly potential enrollees. The one exception is Had Active Ticket (within 90 days) 2, which has the
expected positive sign but only achieves significance at the 0.19 level based on a two-tailed test in
the potential enrollee group. Note, however, that these results for the test sample overstate true
significance levels since they do not correct for the many other exploratory regressions run with
alternative sets of explanatory variables included in them.
In contrast, p-values for the results in the validation sample, shown on the right side of Table 3-1,
are not biased downward since exploratory regressions did not precede tests on the validation
sample. Not surprisingly, these p-values are not consistently as low as those in the test sample are.
The variables that are least significant (i.e., have the highest p-values) in the validation sample are the
socioeconomic variables from the census tract data (Some College and Median Income), age (and age2),
and the variables reflecting exposure to the recruitment variables (Time to Recruit and Time to
Recruit2) particularly in the regression for potential enrollees in the validation sample. Three nonsignificant variables relating to prior work-related activities include the following: (1) the variable for
an active Ticket to Work (TTW) in the 90 days prior to the beneficiary’s recruitment date (Had Active
Ticket w/in 90 days), (2) the variable for a trial work period ending between 5 and 10 years prior to
the recruitment date (Trial Work End Date 5-10 yrs. ago), and (3) the square root of gross earnings
plus net self-employment earnings reported to SSA summed over months 7 to 23 prior to the
recruitment date (Sq. root of Reported Earnings 7-23 mos. pre-recruitment).
1
One cannot directly interpret each reported coefficient as a marginal (ceteris paribus) effect on the probability of enrolling in the
MHTS with the logistic functional form. However, a good approximation to this marginal effect, MEi for the ith X variable, is
computed from the expression MEi=bi x q x (1-q), where bi is the reported coefficient and q is the fraction of the beneficiaries in the
regression sample who in fact enrolled.
2
Had Active Ticket (within 90 days) refers to beneficiaries who activated their Ticket in the Ticket To Work (TTW) program, see
http://ssa.gov/work/aboutticket.html.
Chapter 3: Enrollment and Participation
3-8
Apart from the differences mentioned above, the general pattern of results for the validation
samples is very similar to that for the test sample. Demographic variables such as gender and race were
very strong predictors of enrollment outcome. Beneficiaries on the SSDI rolls for a longer time
(Months on Rolls), beneficiaries receiving both SSDI and Supplemental Security Income (SSI), and
those having a representative payee (Repayee) were much less likely to enroll. As expected, distance
from the beneficiary’s residence to the MHTS site (Ln of Distance from Site) was a strong deterrent to
enrollment. The general pattern of results for the work and work-related activity variables indicated
that holding other factors constant, beneficiaries with a very recent history of work activity (e.g.,
Trial Work End Date 0-3 yrs. post-recruitment date) were less likely to enroll; this was also true for those
with no prior history of an active TTW (Had Active Ticket ever) or a trial work period (Trial Work End
Date 0-3 yrs. post-recruitment date).
Probabilities of enrollment. Table 3-2 shows comparisons of the characteristics of beneficiaries
who were more likely to enroll compared to those who were less likely to enroll according to the
regression model for the validation sample of potential enrollees. The basis for the table is the predicted
enrollment probability derived from the regression model for each beneficiary. Beneficiaries were
grouped into quintiles (and the top decile) based on their predicted enrollment probabilities, and
then mean values of beneficiary characteristics and other explanatory variables are displayed for each
of the quintiles and the top decile. The results for the top decile, in particular, show the
characteristics for those beneficiaries who are estimated to be most likely to enroll in the study.
The mean predicted enrollment probability (in the top row of Table 3-2) ranged from 0.06 for the
lowest quintile to 0.26 for the highest quintile and 0.32 for the highest decile. Comparing mean
characteristics across the quintiles, those in the highest quintile tended to be about 4 years younger
than those in the lowest quintile, were more likely to be male (50% vs. 43%), and more likely to be
black (37% vs. 14%). Those in the highest quintile were also much less likely to have a representative
payee and tended to live closer to the study site than is true for those in the lower quintiles.
With regard to the time pattern of enrollment, beneficiaries in the highest quintile tended to be those
with a longer duration for possible enrollment (Time to Recruit) and those whose contact date was in
2006 or 2007 (rather than 2008).
With regard to SSDI or SSI program history, beneficiaries in the highest quintile were less likely to
be concurrently receiving SSI, and were on the SSDI rolls (Months on Rolls) about 50 months less
time than those in the lowest quintile. Among other recipient and program-related characteristics, a
clear pattern was not observed showing variation among the quintiles among beneficiaries.
Chapter 3: Enrollment and Participation
3-9
However, there did appear to be some tendency for the fraction of beneficiaries not yet enrolled in
Medicare (No Medicare at enrollment) to decrease in the higher quintiles, though these characteristics
were not significant predictors in the exploratory regressions (not shown here). The socioeconomic
variables associated with census tract statistics that were included in the analyses suggested a trend
toward lower education levels and lower median earnings levels for the higher quintiles. All the
work-related activity history variables tended to show increasing work or related activities as the
probability of enrollment increased. Finally, the diagnostic indicators derived from the SSA
administrative data did not show a systematic pattern of variation in relation to enrollment
probability.
Corresponding results for the potential enrollees in the test sample (see Appendix 3C) showed many of
the same patterns seen in Table 3-2, with a few observed differences. The negative trend of
education level for the socioeconomic variables associated with the census tract statistics was
reversed in test sample, as was the negative relationship between enrollment probability and fraction
of the quintile not enrolled in Medicare. The results for the potential enrollees plus possibly potential
enrollees in both the test and validation samples also appear in Appendix 3C.
It is also of interest to compare the means between the beneficiaries in the top decile with
beneficiaries in the top quintile. The most striking differences appear to be in recent work-related
activities in TTW and trial work period experiences. For example, the fraction of beneficiaries with
an active ticket in the 90 days prior to recruitment (Had Active Ticket w/in 90 days) was more than
one-third higher for the highest decile than it was for the highest quintile. This suggests that
targeting MHTS recruitment efforts toward beneficiaries with recent TTW activity would tend to
yield higher enrollment rates.
Characteristics of the Beneficiaries Who Enrolled in the Study
The 23 study sites enrolled 2,238 beneficiaries, with 1,121 randomized to the treatment group and
1,117 randomized to the control group. One of the goals of the randomization process was to
ensure equivalence between the study groups. Thus, a test of the effectiveness of the randomization
process included comparisons on the key demographic and program variables as well as on the
baseline measures of employment, health status, and income. All of the data for these analyses came
from either the administrative records from SSA’s MBR or the MHTS Baseline interview.
Chapter 3: Enrollment and Participation
3-10
Table 3-2.
Mean beneficiary characteristics for quintiles based on predicted enrollment
probability of potential enrollees in the validation sample (N=7,933)
Variable
Enrolled (yes/no)
Age (years)
Gender (Male)
Race (Black)
Rep Payee (yes/no)
Distance from Site (miles)
Months on Rolls
SSI (yes/no)
SSDI<24 mos. (no SSI)
Primary Insured Amount
No Medicare (at enrollment)
Time to Recruit
2006 (Recruitment Yr.)
2007 (Recruitment Yr.)
2008 (Recruitment Yr.)
Some College
Median Income (census tract)
Had Active Ticket (w/in 90
days)
Had Active Ticket (ever)
Trial Work End Date (10 yrs.+
ago)
Trial Work End Date (5-10 yrs.
ago)
Trial Work End Date (0-5 yrs.
ago)
Trial Work End Date (0-3 yrs.
post- recruitment date)
Reported Earnings (1-6 mos.
pre-recruitment)
Reported Earnings (7-23 mos.
pre-recruitment)
No Earnings Report (1-6 mos.
pre-recruitment)
Diagnosis (Affective Disorder)
Primary Diagnosis of Serious
Mental Illness (SMI)
% with Secondary Diagnosis
SMI
% with SMI
Lowest
1st
Quintile
(n=2,530)
0.06
47.86
0.43
0.14
0.32
13.46
148.04
0.25
0.02
8,092.08
0.06
232.10
0.03
0.43
0.54
0.27
26,531.91
2nd
Quintile
(n=1,444)
0.09
46.37
0. 43
0.22
0.16
11.74
116.75
0.18
0.03
8,674.24
0.05
266.45
0.06
0.48
0.46
0.26
25,874.18
3rd
Quintile
(n=1,194)
0.11
46.31
0.43
0.25
0.09
10.90
107.83
0.15
0.02
8,782.19
0.04
279.85
0.07
0.52
0.41
0.26
25,306.46
4th
Quintile
(n=1,254)
0.16
45.35
0.45
0.33
0.05
11.01
97.70
0.15
0.05
8,771.65
0.06
308.35
0.11
0.59
0.30
0.27
25,183.33
Highest
5th
Quintile
(n=1,511)
0.26
43.47
0.50
0.37
0.05
9.37
97.93
0.14
0.03
8,566.14
0.05
348.67
0.19
0.59
0.22
0.25
24,609.61
Top 10%
(n=842)
0.32
43.31
0.49
0.35
0.06
9.53
99.77
0.14
0.02
8,405.18
0.04
377.98
0.23
0.59
0.18
0.23
24,427.10
0.001
0.01
0.02
0.04
0.22
0.29
0.003
0.02
0.04
0.08
0.28
0.37
0.03
0.04
0.07
0.08
0.13
0.12
0.05
0.05
0.04
0.06
0.08
0.08
0.02
0.03
0.04
0.06
0.10
0.12
0.002
0.003
0.003
0.005
0.04
0.05
358.74
193.83
141.20
174.89
275.37
321.75
833.95
529.44
668.16
867.88
1,696.57
2,007.42
0.96
0.95
0.95
0.94
0.83
0.79
0.62
0.70
0.72
0.71
0.68
0.69
0.80
0.72
0.70
0.66
0.71
0.74
0.36
0.34
0.34
0.33
0.35
0.37
0.80
0.72
0.70
0.66
0.71
0.74
Chapter 3: Enrollment and Participation
3-11
Comparisons on demographic and SSDI program variables. Table 3-3 provides a summary of the
demographic and program characteristics of beneficiaries enrolled in the study. The overall mean age
of participants was 47.4 years. The majority of enrolled participants come from the older age range
of 36 to 55 (84.8%). Thus, only a small percentage (15.2%) of the participants was younger (ages 18
to 35). A brief review of the table shows that there were more females than males overall (52.7%) as
well as within each study group. In both groups and overall, 12 percent or fewer were Hispanic. The
majority of the enrollees were white (60.0%) or black (26.3%). In both study groups, approximately
12 percent had less than a high school education. However, the largest group included participants
with some college or some technical school (34.9%), followed by participants with high school
education (26.3%) or a bachelor’s degree (11.6%). Nearly half the enrolled participants (46.2%) were
never married, with 27.4 percent divorced and 16.9 percent married. Among these key demographic
variables, race was the only one to show a significant difference between the treatment and control
groups (p-value = 0.024). On closer inspection, it appears that the difference was associated with the
number of participants indicating “Other” as their race. Of the 209 (9.3%) participants in this
category, 123 (11% of the total) were in the control group and 86 (7.7% of the total) were in the
treatment group.
The SSDI program characteristics of study participants included diagnosis associated with the
primary impairment, insurance status, and length of time on the SSDI rolls. Participants with an
affective disorder comprised approximately 70 percent of the study population, including 68.2
percent of the treatment group and 72.5 percent of the control group. Conversely, 29.7 percent of
the study participants had a diagnosis of schizophrenia, comprising 31.8 percent of the treatment
group and 27.5 percent of the control group. As noted in the table, these differences were
statistically significant (p-value = .024). As a result, analyses reported in later chapters use statistical
control for diagnosis as a covariate.
More than three-quarters of enrollees (76.3%) had been SSDI beneficiaries for more than 2 years
(SSDI > 24 months), indicating that they were eligible for Medicare. Nearly 5 percent were on SSDI
less than 24 months, indicating that they were not yet eligible for Medicare at the time of enrollment
(unless they had a reduced waiting period as SSI recipients). Approximately 19 percent were also on
SSI and likely dual-eligible for Medicare and Medicaid. The average time on the rolls was 103.3
months, with participants in the treatment group having been on the SSDI rolls an average of 106.3
months and those in the control group 103.3 months.
Chapter 3: Enrollment and Participation
3-12
Table 3-3.
Demographic and program characteristics of enrolled beneficiaries by study
condition
Variable
DEMOGRAPHIC CHARACTERISTICS
Treatment
(n=1,121)
Age
Mean
Median
Control
(n=1,117)
years
47.2
49
Full sample
(N=2,238)
years
47.5
49
p-value
years
47.4
49
0.815
Age Group
18-35
36-55
n
176
945
%
15.7
84.3
n
165
952
%
14.8
85.2
N
341
1897
%
15.2
84.8
0.541
Gender
Male
Female
n
519
602
%
46.3
53.7
n
539
578
%
48.3
51.7
N
1058
1180
%
47.3
52.7
0.354
Ethnicity
Hispanic
Not Hispanic
Refused or don’t know
n
121
1000
0
%
10.8
89.2
0.0
n
134
980
3
%
12.0
87.7
0.3
N
255
1980
3
%
11.4
88.5
0.1
0.359
Race
White
Black
Asian
Two or more races
Other
Refused or don’t know
n
674
308
16
33
86
4
%
60.1
27.5
1.4
2.9
7.7
0.4
n
669
281
11
32
123
1
%
59.9
25.2
1.0
2.9
11.0
0.1
N
1343
589
27
65
209
5
%
60.0
26.3
1.2
2.9
9.3
0.2
0.034
Marital Status
Never married
Married
Living as married
Separated
Divorced
Widowed
Refused or don’t know
n
519
196
5
59
312
27
3
%
46.3
17.5
0.4
5.3
27.8
2.4
0.3
n
514
183
19
69
302
29
1
%
46.0
16.4
1.7
6.2
27.0
2.6
0.1
N
1033
379
24
128
614
56
4
%
46.2
16.9
1.1
5.7
27.4
2.5
0.2
0.959
Education
Less than high school
High school or GED
Some college or technical
Associate’s degree
Bachelor’s degree
Some graduate school
Master’s degree
Doctoral degree
Other, refused, or don’t know
n
136
285
402
81
140
20
49
6
2
%
12.1
25.4
35.9
7.2
12.5
1.8
4.4
0.5
0.2
n
135
303
378
99
120
32
35
13
2
%
12.1
27.1
33.8
8.9
10.7
2.9
3.1
1.2
0.2
N
271
588
780
180
260
52
84
19
4
%
12.1
26.3
34.9
8.0
11.6
2.3
3.8
0.8
0.2
0.248
Chapter 3: Enrollment and Participation
3-13
Table 3-3.
Demographic and program characteristics of enrolled beneficiaries by study
condition (continued)
Variable
PROGRAM CHARACTERISTICS
Treatment
(n=1,121)
Control
(n=1,117)
Full sample
(N=2,238)
p-value
Diagnosis
Schizophrenia
Affective disorder
n
357
764
%
31.8
68.2
n
307
810
%
27.5
72.5
N
664
1574
%
29.7
70.3
0.024
Insurance Status
SSDI > = 24mos.
SSDI < 24mos.
SSI
n
857
53
211
%
76.4
4.7
18.8
n
851
54
212
%
76.2
4.8
19.0
N
1708
107
423
%
76.3
4.8
18.9
0.897
Months on SSDI
Mean
106.3
110.3
103.3
0.184
Comparisons on recent employment and health status measures. Beneficiaries enrolled in the
study reported during their Baseline interview if they had worked at any time during the past 2 years.
As shown in Table 3-4, 665 of the 2,238 (29.7%) study participants indicated that they had worked
at some time during the past 2 years. Of these individuals, 316 (28.2%) received random assignments
to the treatment group and the remaining 349 (31.2%) to the control group after completing the
Baseline interview. A Chi-square test indicates there is no significant difference between the
treatment and control groups on this variable.
Table 3-4.
Recent employment history and health status comparisons
Variable
Employment History
Worked in past 2 years
Control
(n=1,117)
N
%
Full sample
(N=2,238)
N
%
p-value
0.114
316
28.2
349
31.2
665
29.7
Mean
Std Dev
Mean
Std Dev
Mean
Std Dev
Physical Health (PCS)
44.2
12.0
44.0
11.9
44.1
11.9
0.673
Mental Health (MCS)
36.3
13.1
36.0
13.0
36.1
13.1
0.597
SF-121
1 N=
Treatment
(n=1,121)
N
%
2,233. Five beneficiaries answered “don’t know” or refused individual SF-12 items. Those scores were dropped from this analysis.
Beneficiaries enrolling in the study also answered questions about their health. Table 3-4 presents
descriptive data for all enrollees including comparisons between the treatment and control groups.
The two groups had almost identical mean scores, thus differences between the groups proved not
to be statistically significant based on a t-test statistic.
Chapter 3: Enrollment and Participation
3-14
Comparison on physical health conditions measures. The screener that preceded the Baseline
interview included items about current medications and physical conditions that might interfere with
employment. The investigators suspect that data from these sources (Table 3-5) are underestimates
of physical health conditions, because at the time of the interview most beneficiaries had limited
documentation of their current medications to report during the interview. However, information
that is more detailed came from the treatment group later during the implementation of the SMM.
Chapter 6 provides more details on physical health conditions from multiple sources for the
treatment group only.
Table 3-5.
Anemia
Physical health conditions identified in CAPI screener
Physical health condition
Autoimmune Disorders
Treatment
(N=1,121)
freq
%
14
1.2
Control
(N=1,117)
freq
%
19
1.7
p-value
0.375
11
1.0
13
1.2
0.675
Blindness
0
0.0
1
0.1
0.499
Brain Damage
5
0.4
5
0.4
1.000
Cancer
11
1.0
9
0.8
0.659
Cardiovascular Diseases
23
2.1
26
2.3
0.656
4
0.4
8
0.7
0.265
84
7.5
119
10.7
0.009
167
14.9
198
17.7
0.070
97
8.7
113
10.1
0.235
143
12.8
186
16.7
0.009
19
1.7
21
1.9
0.741
1
0.1
5
0.4
0.124
Hyperlipidemia
138
12.3
170
15.2
0.046
Hypertension
224
20.0
271
24.3
0.015
Liver Disease
9
0.8
8
0.7
0.813
Migraines
13
1.2
32
2.9
0.004
Narcolepsy
1
0.1
2
0.2
0.624
Neuromuscular/Degenerative Disorders
42
3.7
40
3.6
0.835
Neuropathy
19
1.7
21
1.9
0.741
Renal Disease
19
1.7
30
2.7
0.109
Seizure Disorder
26
2.3
27
2.4
0.879
Thyroid Disorders
85
7.6
88
7.9
0.375
4
0.4
2
0.2
0.675
Chronic Fatigue Syndrome
Chronic Lung Disorder
Chronic Pain Conditions
Diabetes
Gastrointestinal Disorders
HIV
Hearing Loss
Other
Chapter 3: Enrollment and Participation
3-15
The relative frequencies of conditions were similar across groups. On average, the control group had
somewhat higher rates of physical conditions. The t-test revealed that the differences between
treatment and control group beneficiaries were significant for Chronic Lung Disorder (p-value =
0.0092); Gastrointestinal Disorders (p-value = 0.0093); Hyperlipidemia (p-value = 0.0458);
Hypertension (p-value = 0.0147); and Migraines (p-value = 0.0041).
Comparisons on income measures. Both the Baseline and Followup interviews collected data
about various sources of income. The sources assessed included the following:

Paid employment;

SSDI;

SSI;

Social Security Retirement or Survivor’s Benefits;

VA or other armed services disability benefits;

State or county social welfare benefits (e.g., general assistance or public aid, food stamps
or assistance from the Temporary Assistance for Needy Families program);

Vocational programs (e.g., vocational rehabilitation, Job Training Partnership Act,
Easter Seals);

Unemployment benefits;

“Regular payments” from retirement, pension, investing or savings income, alimony and
child support;

Money (including loans, gifts, or bill payments) from family members;

“Other income” (from informal sector jobs); and

Total household income (not asked if living alone or if in residential housing with staff
or other clients).
In general, the analyses of each measure found little difference (not shown) between the treatment
and control groups on most of the variables studied, and no statistically significant differences.
Table 3-6 presents the average levels of individual income of beneficiaries enrolled in the study
reported by treatment and control participants during the Baseline interview. There were no
Chapter 3: Enrollment and Participation
3-16
statistically significant differences between the treatment and control groups on the “total individual
income” outcome variable.
Table 3-6.
Total individual income1
Variable
All responses
Treatment
Mean
SE
N
1055.5 18.5 1121
Control
Mean
SE
N
1015.0 14.7 1117
Total sample
Mean
SE
N
1035.3 11.8 2238
All non-missing
1064.7
18.7 1090
1032.6
14.8 1076
1048.7
11.9
2166
0.15
Non-HH income
1058.4
32.2
19.7
1026.6
18.7
918
0.11
451
996.0
467
p-value2
1
This is the sum of all individual income items. The figures for “All responses” recodes “Don’t know” or “Refused” responses for
individual items as zeros. The figures for “All non-missing” only include responses for persons who did not report any “DK” or “Refused”
responses for any individual items. The “Non-HH income” figures are for all persons who (by virtue of their living situations) were not
asked to report household income.
2
The reported p-value is the two-tailed p-value of the coefficient of the treatment dummy from a bivariate regression assuming
clustering of errors by site.
Why Did Eligible Beneficiaries Refuse to Enroll?
The study collected some information about the reason a beneficiary refused to engage in a RIG
meeting or enroll after attending one or more RIG meetings. For each beneficiary who refused
participation, the RA recorded in the SMS (with varying levels of detail) the reason for the refusal.
The investigative team developed a categorical list of refusal reasons, with assistance from RAs from
across the study sites. The resulting reasons in their order of magnitude appear in Table 3-7. The
table shows both all beneficiary refusals, as well as refusals from those who attended one or more
RIG meetings (a subgroup of all beneficiary refusals).
Table 3-7.
Refusal frequencies
All
Refusal category
General disinterest
Frequency
(n=8103)
3029
Percent of
total
%
37.4
Attended RIG
Percent of
Frequency
total
(n=684)
%
330
48.3
Work-related (i.e., cannot work)
1728
21.3
87
12.7
Physical health
1242
15.3
51
7.5
Other
1049
12.9
77
11.3
Life issues or other obligations
586
7.2
93
13.6
Symptom-related
392
4.8
22
3.2
Unknown
63
0.8
10
1.5
Missing
14
0.2
14
2.1
Chapter 3: Enrollment and Participation
3-17
0.07
As presented in Table 3-7, the majority of refusals were due to general disinterest for all beneficiaries
(37.4%) and for those who attended one or more RIG meetings (48.3%). For all beneficiaries, other
common reasons for refusing to enroll in the MHTS included work-related reasons (21.3%),
physical health (15.3%), and other reasons (e.g., the beneficiary was moving, or someone else such as
a spouse or parent refused on behalf of the beneficiary, 12.9%). Less common reasons included life
issues or obligations (e.g., taking care of a sick parent or family member, 7.2%) and symptom-related
reasons (4.8%). For those beneficiaries who attended one or more RIG meetings, other common
reasons for refusing to enroll in the MHTS included life issues or obligations (13.6%), work-related
reasons (12.7%), and other reasons (11.3%). Less common reasons included physical health
(7.5%) and symptom-related reasons (3.2%).
Discussion
This chapter provides answers to the many questions about enrollment. However, in doing so, it has
also identified a number of important issues for the study. First, it is important to know that the
randomization process was successful with regard to the key measures at baseline. None of them—
employment, income, or health status—showed a significant difference between the treatment and
control groups at baseline. There were, however, some differences between the treatment and
control groups on other variables. The research found significant differences between the treatment
and control participants on race, diagnosis, and five of the self-reported health conditions (including
hypertension, chronic lung disorder, gastrointestinal disorders, hyperlipidemia, and migraines—all
higher among the control group participants). The treatment group included a slightly higher
percentage of blacks, while the control group included a higher percentage of participants in the
“Other” race category.
With regard to diagnosis, the treatment group showed a higher percentage of participants with
schizophrenia, while the control group showed a higher percentage of participants with affective
disorder. Prior to recruitment, the investigators decided not to stratify using diagnosis due to the
large number of beneficiaries with these disorders. The important stratification variable was
insurance status, as the need to ensure equal distribution between the two groups of the low
numbers of beneficiaries with less than 24 months on the rolls, and dual-eligible beneficiaries. With
61,530 beneficiaries and a ratio of approximately 3 to 1 beneficiaries with affective disorders relative
to those with schizophrenia, there did not seem to be a need to stratify on the diagnosis variable.
Chapter 3: Enrollment and Participation
3-18
The difference between the groups in the number of participants reporting health conditions was
not necessarily problematic. The overall numbers were not high relative to the large numbers of
participants in the treatment and control groups. In addition, there were some concerns about the
quality of self-reported health conditions in the Baseline interview. Many of the beneficiaries did not
bring their medications with them to the interview as directed. Those who did bring their
medications were able to provide better information on their health conditions. In addition, there
were other indicators of health collected on treatment group participants during the study. As
reported in Chapter 6, there were many more health conditions than originally reported in the
Baseline interview by these beneficiaries. Thus, the baseline reporting is likely an underestimate of
health conditions.
The most interesting conclusion suggested from the analysis of participation was that data routinely
available in SSA’s administrative records were predictive of participation in the MHTS intervention.
These variables included presence of a representative payee, distance from study site, months on
SSDI, and recent TTW activity. The differences in the validation sample, between the predicted
probability for the highest quintile (26% for potential enrollees) and the lower quintiles in Table 3-2,
highlight this conclusion. More generally, the figures in Tables 3-2 and the corresponding tables in
Appendix 3C suggest that a participation rate in excess of 25 percent might be a reasonable
expectation with recruitment efforts focused on a targeted group (e.g., beneficiaries with a recent
active TTW).
However, without targeting particular beneficiaries, a reasonable rate of enrollment might be closer
to 14 percent. The analyses found that 26 percent of the top 20 percent of the potential enrollee
group were predicted to enroll, resulting in 5.2 percent of the group. For the remaining 80 percent
of potential enrollees (i.e., the lower four quartiles), the corresponding figures were as follows: 1.2
percent (.06 x .20), 1.8 percent (.09 x .20), 2.2 percent (.11 x .20), and 3.2 percent (.16 x .20).
Therefore, a total predicted enrollment rate would be 13.6 percent. By implication, one might
expect an enrollment of 306,000 SSDI beneficiaries out of the 2.25 million who have a psychiatric
impairment.
One other qualification to these projections should also be borne in mind. As noted earlier, these
analyses were for a randomized trial rather than for an expanded implementation effort (where
researchers assign all enrollees to the “treatment” group). It seems plausible that the random
assignment in MHTS trial may in fact have led to a lower participation rate than would have been
observed in an expanded implementation process; however, this is unknown.
Chapter 3: Enrollment and Participation
3-19
Beneficiaries’ concerns about returning to work or physical health issues accounted for nearly 40
percent of the documented reasons for refusal. However, many of the enrolled treatment group
participants experienced anxieties about work. In some cases, working with the SE specialist
alleviated these fears. In addition, the Nurse Care Coordinator spent considerable time working with
beneficiaries to address physical health challenges. Thus, it is possible that with further education
and knowledge of how the intervention could address personal challenges, the uptake rate would
have been higher.
Chapter 3: Enrollment and Participation
3-20
Chapter 4
Outcomes
This chapter presents analyses and results that answer the fundamental research question asked of
the Mental Health Treatment Study (MHTS): To what extent does delivering appropriate mental
health treatment and employment supports lead to better employment, health status, and quality of
life among Social Security Disability Insurance (SSDI) beneficiaries with schizophrenia or an
affective disorder (SSA, 2005, p. 6)? The answer to this question comes from a direct head-to-head
test between the outcomes attained by the treatment group and the control group.
This chapter is divided into three sections, including employment outcomes, earnings and income,
and health and quality of life. The Employment Outcomes section begins with an assessment of the
primary study outcome—employment rate. This section also contains a large number of additional,
secondary analyses of other employment-related measures (e.g., months employed, average weekly
earnings, average hours at main job, highest hourly wage, satisfaction with job at study exit). All of
these analyses include tests of differences between key subgroups of interest to the Social Security
Administration (SSA). Summary tables indicating the subgroup comparisons for which the results
were statistically significant appear in the chapter; however, to save space and facilitate reading of
the chapter, the descriptive statistics and test data appear in an appendix. The Employment
Outcomes section also includes analyses associated with the concept of a steady worker (referring to
participants who reported employment in 10 or more months during the study) and types of jobs
obtained by study participants. A key aspect of these secondary analyses includes several multivariate
analyses that attempt to predict employment. The section ends with presentations of data suggesting
why some study participants worked part-time and others did not work at all.
The section on Earnings and Income includes a large number of secondary outcome analyses. The
first analyses concern reports of past month‘s earnings (averaged over the eight post-baseline
interviews), and comparisons between the reported earnings and Substantial Gainful Activity (SGA).
The section also includes comparisons between treatment and control group participants‘ income
from sources other than earnings, including individual and household income and trends in earnings
differentials. This section also reports the estimates of MHTS impacts on earnings and income.
The section on Health and Quality of Life presents assessments of the remaining primary study
outcomes. These non-vocational outcomes include mental health status, physical health status, and
Chapter 4: Outcomes
4-1
quality of life. The section includes a number of secondary tests of differences between key
subgroups, including age, gender, diagnosis (according to SSA administrative files), and education.
Additionally, the section includes several multivariate analyses that attempt to predict mental health
status and quality of life. The final section includes a secondary outcome analysis of the study impact
on alcohol and drug abuse as measured by the Addiction Severity Index.
Multiple comparisons. This chapter includes a large number of statistical tests, including many ttests and other comparisons. In most cases, these multiple tests are all different ways of looking at
the basic hypothesis that the treatment intervention improved the primary outcomes of
employment, health, and quality of life. The employment outcomes, particularly, are measured in
different ways, such as whether employed, amount of earnings, hours worked, and income, all of
which are related. There are many points of view on the issue of multiple comparisons (e.g.,
Proschan and Waclawiw 2000, Savitz and Oshan 1998, Thompson 1998, Thompson 1997, Savitz
and Olshan 1995, Rothman 1990, and O‘Brien 1983). In his excellent article on the subject,
Schochet (2008) reviews a number of issues related to this question, such as whether analyses are
confirmatory or exploratory, whether the study can focus on a few rather than many outcomes, and
whether outcomes can be divided into domains.
In the sense of Schochet (2008), this complex study involves both confirmatory and exploratory
hypotheses. The general confirmatory hypothesis is that the intervention improves the primary
outcomes. However, the question of how the intervention improves outcomes is complex, involving
many different measures, and is largely, exploratory. In this study, it is not appropriate to focus on a
few measures, which is one of the strategies suggested by Schochet, since each measure shows a
slightly different picture of one of the main overall study endpoints of ―improved employment
outcomes.‖
In the classic multiple comparisons scenario, one conducts a series of independent hypothesis tests;
in this situation, the Type I error is indeed quite inflated. However, the approach here is to look at
these measures as a whole, as opposed to looking at individual measures for the primary outcome of
employment. While it is true that the Type I error for any given individual test is inflated, and that
individual tests are discussed, the emphasis is on interpreting results across measures. Moreover, the
fact remains that nearly all measures examined indicate improved outcomes for the treatment group,
generally with highly significant p-values. The reader is encouraged to assess the results presented
here in the broader context of the four primary outcomes defined at the outset of this study by
SSA—employment, health status (physical and mental), and quality of life.
Chapter 4: Outcomes
4-2
Employment Outcomes
Employment rate. Employment rate was the primary outcome for the study. The policy interest is
whether the treatment intervention offered participants a better chance of obtaining employment
than did the current system of ―services as usual‖ (as represented by the control group). Table 4-1
shows the employment rates for any job and competitive jobs by treatment and control group. The
treatment group attained significantly better employment rates than the control group both in terms
of obtaining any type of job and for competitive jobs. The data represent participants who reported
having a job at any time during their 24 months of study participation. These unduplicated numbers
count participants only once, including those who worked more than one month, or more than one
job. The differences for both types of jobs of approximately 20 percentage points were statistically
significant and substantively meaningful from both a clinical and social policy perspective.
Table 4-1.
Employment rates by treatment and control group
Variable
Employment rate (any)
Employment rate (competitive)
Treatment
(N=1,004)
n
%
605
60.5
526
52.6
Control
(N=1,051)
n
%
423
40.3
347
33.0
p-value
<0.001
<0.001
NOTE: Weighted percents may not be consistent with unweighted counts.
Figure 4-1 shows a comparison of the monthly employment rates (any employment) for the
treatment and control groups. The first 6 months of the study shows a sharp rise in employment
rates across both groups, ranging from approximately 5 to 6 percent in month 1 to between 14 and
18 percent in month 6. The average employment rate over the 6-month period was 14 percent for
the treatment group and 11 percent for the control group. Between months 7 and 12, these rates
increased to 26 percent and 16 percent for the treatment and control groups respectively. The
employment rates for the control group remained steady at 16 percent (on average) for the
remainder of the study. However, the employment rates for the treatment group continued to rise to
30 percent and remained there throughout the second half of the study. As shown by the standard
error bars on the graph, the differences between the treatment and control groups were not
significant throughout the first three months of the study. At month 4, the differences became
significant and continued to widen at a rate of 1 to 2 percent in any given month until reaching a
plateau of over 15 percent at month 16. From then to the end of the study, the treatment group
continued its advantage over the control group by 13 to 16 percentage points.
Chapter 4: Outcomes
4-3
Figure 4-1.
Monthly employment rates for any job by treatment and control group over the 24month study period
50%
45%
Treatment
40 %
Control
35%
30%
25%
20%
15%
10 %
5%
0%
1
2
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Months
Figure 4-2 presents employment rates by study site (see Appendix 4A for statistics). Note that this
graph presents overall employment rates, by study site, over the 24-month study period (not
monthly). Thus, the employment rates appearing in Figure 4-2 are much higher than the monthly
rates that appear in Figure 4-1. The graph shows substantial variation in the employment rates by
site. Over the 24-month study period, employment rates for the treatment group ranged from a low
of 34 percent (site 21) to a high of 77 percent (site 15) with a standard deviation of 11 percent.
Commensurate rates for the control group were 23 percent (site 21) and 60 percent (site 3) with a
standard deviation of 10 percent. Thus, the amount of variability is similar across both treatment
and control groups.
Chapter 4: Outcomes
4-4
Figure 4-2.
Employment rates for any employment by treatment group and control group and
by study site
80%
Treatment
Control
70%
60%
50%
40%
30%
20%
10%
0%
01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Site
Table 4-2 presents employment rates for both any job and competitive jobs by age, gender,
diagnosis, and education.1 All of these stratifications showed significant and meaningful differences.
In general, participants with different demographic and clinical characteristics benefitted significantly
and meaningfully from the treatment. For example, among participants diagnosed with an affective
disorder, the treatment group demonstrated significantly better employment rates for both any job
and for competitive jobs than control group. While the differences in employment rates for any job
and competitive jobs were smaller for participants with schizophrenia, they too were significant and
meaningful.
1
Note that competitive employment is a subgroup of any employment, so participants included in statistics for competitive employment are also
counted in statistics for any employment.
Chapter 4: Outcomes
4-5
Table 4-2.
Employment rates by treatment and control group, stratified by age, gender,
diagnosis, and education
Variable
Employment rate (any)
Age: 18 to 34
Age: 35+
Gender: Male
Gender: Female
Diagnosis: Affective Disorder
Diagnosis: Schizophrenia
Education: Less than HS
Education: HS graduate
Education: More than HS
Treatment
(N=1,004)
n
%
605
60.5
61
70.9
544
59.5
285
61.6
320
59.5
417
60.3
188
60.9
78
65.1
151
59.2
376
60.2
Control
(N=1,051)
n
%
423
40.3
40
47.2
383
39.7
199
39.6
224
41.0
305
39.5
118
42.5
49
38.2
110
38.1
264
41.7
p-value
<0.001
0.002
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
Employment rate (competitive)
Age: 18 to 34
Age: 35+
Gender: Male
Gender: Female
Diagnosis: Affective Disorder
Diagnosis: Schizophrenia
Education: Less than HS
Education: HS graduate
Education: More than HS
526
59
467
251
275
371
155
67
140
319
347
33
314
158
189
253
94
37
86
224
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
52.6
68.6
51.1
54.3
51.1
53.7
50.2
56.3
55.1
50.9
33.0
38.9
32.5
31.2
34.6
32.7
33.8
28.7
29.9
35.3
NOTE: Weighted percents may not be consistent with unweighted counts.
Other employment outcomes. Table 4-3 presents a summary of other employment outcomes. The
table shows employment outcomes for three groups of participants: all participants in the study,
participants who worked at least one job (any job), and participants who worked at least one
competitive job. The third group, participants who worked a competitive job, is a subset of the
second group who worked at least one job. Further, the second group is a subset of all participants
in the study. The distinctions among the three groups warrant a brief explanation. First, many
people believe that type of employment is important. The Individual Placement and Support (IPS)
model for supported employment (SE) services gives priority to competitive jobs as important
aspects of community integration and patient-centeredness, but other vocational programs are more
likely to use sheltered or non-competitive jobs. People with psychiatric illness themselves clearly
prefer competitive jobs, which by definition are in integrated rather than segregated work settings,
because they value community integration (Bedell, Draving, Parrish, Gervey, & Guastadisegni,
1998). The U.S. Rehabilitation Act of 1973 (Pub. L. No. 93-112, 87 Stat. 394 [September 26, 1973])
Chapter 4: Outcomes
4-6
also emphasizes employment in integrated work settings as part of the American commitment to
social inclusion of people with disabilities.
Table 4-3.
Other employment outcomes for all participants, participants who worked at least
one job, and participants who worked at least one competitive job by treatment and
control group
Variables
All participants
Months to 1st job
Total months employed
Consecutive months of employment at
study exit
Average weekly earnings at main job
Average hours per week at main job
Highest hourly wage
Participants who worked at least 1 job
Months to 1st job
Total months employed
Consecutive months of employment at
study exit
Average weekly earnings at main job
Average hours per week at main job
Highest hourly wage
Job satisfaction with main job at study end1
Participants who worked at least 1
competitive job
Months to 1st job
Total number of months employed
Consecutive months of employment at
study exit
Average weekly earnings at main job
Average hours per week at main job
Highest hourly wage
Job satisfaction with main job at study end2
Treatment
M
SD
N=1,004
†
†
6.23
7.48
3.22
6.55
116.58
11.93
7.65
139.30
12.47
8.77
n=605
7.37
6.17
10.30
7.09
5.32
7.72
192.69
19.72
12.65
38.77
131.70
10.07
7.98
8.34
n=526
Control
M
SD
N=1,051
†
†
3.65
6.44
1.79
5.27
d
p-value
†
2.58
1.43
†
<0.001
<0.001
140.59
11.20
8.12
40.54
4.29
2.56
<0.001
<0.001
<0.001
n=423
6.68
6.31
9.04
7.33
4.45
7.55
0.69
1.26
0.87
0.016
0.001
0.005
4.07
0.76
0.02
1.49
0.080
0.173
0.443
0.023
76.04
7.64
5.09
188.62
18.96
12.63
37.28
166.91
9.82
8.25
8.44
n=347
7.74
9.26
4.27
6.04
6.66
6.94
7.22
8.35
3.72
6.16
6.82
6.86
0.52
0.91
0.55
0.107
0.017
0.113
201.11
20.49
11.36
38.40
143.61
10.69
5.90
8.27
193.32
19.40
11.54
37.43
176.16
10.14
7.02
8.95
7.79
1.09
-0.18
0.97
0.060
0.097
0.645
0.347
† Not applicable.
1The Ns for job satisfaction are n=335 for the treatment group and n=178 for the control group because this measure includes only
participants who reported working in a job during the final Followup interview (whereas the other measures include participants who
reported working in a job during any of the eight post-baseline interviews).
2The
Ns for job satisfaction are n=251 for the treatment group and n=138 for the control group because this measure includes only
participants who reported working in a competitive job during the final Followup interview (whereas the other measures include
participants who reported working in a competitive job during any of the eight post-baseline interviews).
Second, separate analyses restricted to those who obtain employment only must be interpreted in
recognition of the fact that interventions that obtain higher rates of employment are including as
Chapter 4: Outcomes
4-7
workers people who have less motivation, readiness, background, skills, or other characteristics as
workers. Other things being equal, they should therefore have relatively poor ―other employment
outcomes‖ compared to programs that focus on clients with greater motivation, readiness,
background, skills, or other characteristics as workers. Thus, it is important to measure these other
employment outcomes (such as total months employed, average weekly earnings, and consecutive
months employed) to further assess the effectiveness of the treatment intervention.
In the context of the entire study sample, the other employment outcomes for the treatment group
were statistically and meaningfully significant over those of the control group. As noted in Table 4-3,
the treatment group (as a whole) was employed for more months over the study period (6.23 vs.
3.25, p-value < 0.001), employed longer as measured by consecutive months employed at study exit
(3.22 vs. 1.79, p-value < 0.001), attained higher average weekly earnings at their main job ($116.58 vs.
$76.04, p-value < 0.001), worked more hours per week (11.93 vs. 7.64, p-value < 0.001), and had
higher hourly wages ($7.65 vs. $5.09, p-value < 0.001).
In the more specific contexts of participants who obtained at least one job or participants who
obtained at least one competitive job, the treatment group demonstrated advantages on some but
not all of the other employment outcomes. Time to first job, which of course applies only to those
who obtained a job, showed an advantage for the control group participants, in relation to any job
but not competitive jobs. This finding suggests that some control group participants went to work
rapidly in non-competitive jobs.
The treatment group exceeded the control group on total months employed in any job (10.30 vs.
9.04, p-value = 0.001) and in competitive employment (9.26 vs. 8.35, p-value = 0.017); these
differences were also statistically significant when analyses were restricted to workers only.
Consecutive months of employment among workers at study exit favored the treatment group for
both any employment (5.32 vs. 4.45, p-value = 0.005) and competitive employment (4.27 vs. 3.72,
p-value = 0.113). These results were significant for any employment, but not significant for
competitive employment. The 24-month end date censored this outcome, which quite possibly
could become more significant over time. Previous IPS followup studies have shown greatly
extended job tenure over time (Becker, Whitley, Bailey, & Drake [2007]; Salyers, Becker, Drake,
Torrey, & Wyzik [2004]).
Average weekly earnings among workers favored the treatment group for both types of
employment, but results were only marginally significant. Most of these participants were in the early
Chapter 4: Outcomes
4-8
phases of returning to work, perhaps anticipating increased earnings in the future, as they expanded
their work hours.
Average hours of work per week at the main job were non-significantly greater for the treatment
group for both any employment and competitive employment. These workers were in the early
phase of returning to work and may have felt constrained by fears regarding stress, Social Security
regulations, or other factors.
Highest hourly wage did not differ between groups for any employment or competitive
employment. Again, participants in both conditions were in early phases of returning to work.
Satisfaction with main job at study exit significantly favored the treatment group among participants
with any job, and was non-significant among those with competitive jobs.
Other employment outcomes by subgroups. For each of the seven other employment outcomes,
Table 4-4 presents a summary of the results of t-tests that were conducted to determine whether
differences in mean values between treatment and control group participants were statistically
significant, for the following subgroups: age, gender, diagnosis, and education. These comparisons
are presented for the three participant groups described in the previous section—all participants in
the study, participants who worked at least one (any) job, and participants who worked at least one
competitive job. Summary descriptive statistics and the results of the significance tests are presented
in Appendix 4B for all measures and all subgroups shown in Table 4-4.
In general, the pattern of results is similar to that described in the previous section, when comparing
treatment group to control group participants overall. Participants with different demographic and
clinical characteristics benefitted from the treatment intervention in the context of the entire study
sample (i.e., all participants), regardless of age, gender, diagnosis, or education status. In other words,
the other employment outcomes for the treatment group were statistically and meaningfully
significant over those of the control group. More specifically, treatment group participants were
employed for more months; had more consecutive months of employment at study exit; averaged
higher weekly earnings; averaged more hours per week; and had higher hourly wages compared to
control group participants for all subgroups, with one relatively minor exception. Although younger
treatment group participants between the ages of 18 and 34 had more consecutive months of
employment at study exit than control group participants in the same age group (3.30 vs. 2.38,
p-value = 0.9305) this difference was not statistically significant, more than likely because of the
small N for this younger age group.
Chapter 4: Outcomes
4-9
Table 4-4.
Statistically significant subgroup comparisons for other employment outcomes
including all participants, participants who worked at least one job, and participants
who worked at least one competitive job by age, gender, diagnosis, and education
Variables
All participants
Total months employed
Age
18 to 35 35+
Consecutive months of
employment at study exit
Average weekly earnings at main
job
Average hours per week at main
job
Highest hourly wage
Participants who worked at least 1 job
Months to 1st job
Total months employed
Consecutive months of
employment at study exit
Average weekly earnings at main
job
Average hours per week at main
job
Highest hourly wage
Job satisfaction with main job at
study end
*
Diagnosis
A
S
Education
< HS HS grad > HS
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Participants who worked at least 1 competitive job
Months to 1st job
Total number of months employed
*
Consecutive months of
employment at study exit
Average weekly earnings at main
job
Average hours per week at main
job
Highest hourly wage
Job satisfaction with main job at
study end
Gender
M
F
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
p-value < 0.05
When analyzing these same employment outcomes for participants who obtained at least one job or
participants who obtained at least one competitive job, the treatment group demonstrated
advantages on some, but not all, outcomes for the various subgroups presented. As shown in
Chapter 4: Outcomes
4-10
Table 4-4, there were statistically significant differences between treatment and control for months to
first job; however, examination of Appendix 4B reveals that the direction of the test favored control
group participants. Participants randomized to the control group went to work more rapidly than
participants randomized to the treatment group for participants aged 35 or older, females, and those
diagnosed with schizophrenia. When restricting months to first job even further, to participants who
worked at least one competitive job, observed differences between the treatment and control group
for all subgroups presented were not statistically significant.
When reviewing the entire sample, the treatment group reported working more months (i.e., total
months employed) during the 24-month study period than the control group for all subgroups, and
these differences were statistically significant. However, for participants who worked at least one
job, the treatment group showed an advantage over the control group for only older participants,
males, those diagnosed with an affective disorder, and participants with more than a high school
diploma. These same subgroups also showed a statistically significant advantage for the treatment
group when restricting the sample to participants who worked at least one competitive job.
Average weekly earnings were statistically significant for all subgroups when comparing differences
between treatment and control group participants for all participants (regardless of their
employment status). However, fewer significant differences were observed between treatment and
control participants when restricting the comparison to participants who worked at least one job (for
both any employment and competitive employment). This pattern of fewer significant observed
differences when comparing treatment and control participants among working participants only
holds true for average hours worked per week as well. In fact, for average hours worked per week,
when restricting the comparison to participants who worked at least one competitive job, statistically
significant differences occur only for females and participants with an affective disorder.
While statistically significant differences emerge for highest hourly wage among all subgroups in the
context of all participants, there were no significant differences between the treatment and control
groups when analyzing workers. Satisfaction with main job at study exit, for those who worked at
least one job, significantly favored the treatment group for older participants, females, those with an
affective disorder, and those with higher education. When looking at competitive jobs only, there
were no statistically significant differences between treatment and control participants for any of the
subgroups analyzed.
Steady workers. Classification of study participants into three work categories allowed for
examination of the characteristics of participants who successfully returned to work. After careful
Chapter 4: Outcomes
4-11
examination of the MHTS data and the long-term employment followup literature, the criterion for
a ―steady worker‖ was set at a minimum of 10 months of employment during the 24-month study
period. This criterion is conceptually similar, if not more stringent, than the definition of steady
worker of ―employed at least 50 percent of the time,‖ which has been used in the literature (Becker,
Whitley, Bailey & Drake, 2007; Salyers, Becker, Drake, Torrey & Wyzik, 2004).2 Additional cut
points were set at 3 months and less than 10 months to identify the ―erratic worker,‖ and 3 months
or less to identify the ―minimal worker (including no work).‖
Table 4-5 presents the distribution of work status of participants by study arm. Findings indicate
that there are more participants in the steady worker category among the treatment group
participants (30% vs. 16%) and there were more participants in the minimal worker category among
the control group participants (53% vs. 72%). These differences between treatment and control
groups were statistically significant (p-value < 0.001).
Table 4-5.
Classification of worker types by treatment and control group
Variable
Steady worker
Erratic worker
Minimal worker (including no work)
Treatment
(N=1,004)
n
%
295
29.5
175
17.5
534
53.0
Control
(N=1,051)
n
%
166
15.8
132
12.6
753
71.6
p-value
<0.001
NOTE: Weighted percents may not be consistent with unweighted counts.
Significantly more treatment group participants comprised the steady worker category within each of
the key demographic categories of age, gender, diagnosis, education, and the indicator of having
worked in the past two years (at baseline). Table 4-6 presents the distributions by treatment and
control for the two study age groups of interest, gender, diagnosis, and education.
2
As seen in Table 4-3, workers needed an average of seven months to obtain their first job. Assuming it took a working participant on average seven
months to obtain a first job, he or she would then have 17 months left remaining in the 24-month study period. In order to meet the definition of
steady worker (i.e., employed at least 50% of the time), a participant would have to report working at least half of that 17 months (i.e., 8.5 months).
So the MHTS criterion of 10 months of employment during the 24-month study period is conceptually similar to that described in the literature,
taking into account the amount of time it takes for an MHTS beneficiary to obtain a first job.
Chapter 4: Outcomes
4-12
Table 4-6.
Characteristics of steady workers by treatment and control group
Variable
Age: 18 to 34
Age: 35+
Gender: Male
Gender: Female
Diagnosis: Affective Disorder
Diagnosis: Schizophrenia
Education: Less than HS
Education: HS graduate
Education: More than HS
Worked in past 2 years (baseline)
Not worked in past 2 years (baseline)
Treatment
n/N
25/86
270/918
142/465
153/539
213/693
82/311
35/120
68/256
192/628
123/285
172/719
%
29.6
29.5
30.7
28.5
30.9
26.5
28.7
26.7
30.8
43.1
24.2
Control
n/N
13/85
153/966
67/503
99/548
115/772
51/279
17/127
46/287
103/637
98/330
68/721
%
15.5
15.8
13.2
18.2
15.0
18.2
13.1
16.0
16.3
29.9
9.4
p-value
0.031
<0.001
<0.001
0.001
<0.001
0.017
0.003
0.003
<0.001
<0.001
<0.001
NOTE: Weighted percents may not be consistent with unweighted counts.
Types of occupations obtained. The interviewer recorded job titles verbatim during each
followup interview. This section examines those data and the types of occupations that participants
held during the study. Classification of the occupational types followed the Standard Occupational
Classification (SOC) System maintained by the U.S. Bureau of Labor Statistics (BLS)
(http://www.bls.gov/soc/), which includes an online search engine matching a large database of job
titles to 6-digit SOC codes. The 2010 SOC System is a standard method of classifying workers‘ jobs.
It is widely used across many Federal statistical agencies to document the types of jobs (occupations)
and numbers of workers doing them. The six digit code level includes 840 detailed occupations.
These are often aggregated into smaller groupings for particular users. The standard grouping
includes 461 broad occupations, 97 minor groups, and 23 major groups.
For users who wish to present occupational data at an even broader level, the 2010 SOC User Guide
(http://www.bls.gov/soc/soc_2010_user_guide.pdf) has recommended a high-level aggregation of
the 23 major groups into 6 categories. These six aggregated occupational categories include:

Management, business, science, and arts occupations;

Service occupations;

Sales and office occupations;

Natural resources, construction, and maintenance occupations;

Production, transportation, and material moving occupations; and

Military specific occupations.
Chapter 4: Outcomes
4-13
Table 4-7 shows a comparison of the monthly frequency of jobs held in each of these SOC
categories between the treatment and control groups, with the exception of military specific
occupations; none of the MHTS participants obtained military specific jobs during the study.
Therefore, all tables related to type of job report on the remaining five high-level aggregated
categories. Summarized at the bottom of Table 4-7 are the mean frequencies for jobs reported
during the Baseline interview (i.e., jobs worked the past 2 years prior to baseline), Year 2 (the second
12 months of the 24-month study period), and 2-Year Total (months 1-253). The meaningful statistic
is the percentage of participants within each SOC category. The process involved summing the
numbers within each group, then calculating the percentage from these totals.
At baseline, the distributions of new study participants across the five occupational categories were
remarkably similar between treatment and control groups. A Chi-square test of differences between
treatment and control resulted in no statistically significant differences. Service occupations and sales
and office occupations represented nearly 70 percent of all jobs held by either treatment (67.2%) or
control (68.4%) participants. About half of those percentages were represented by management,
business, science, and arts occupations and production, transportation, and material moving
occupations, evenly spread across treatment (30.1%) and control (28.3%) group participants. Less
than three percent of treatment or control group participants held occupations in the natural
resources, construction, and maintenance category.
The results for Year 2 were similar to those for the 2-Year Total. The Year 2 results may be more
credible in that the employment rates were low during the first several months after study
enrollment. In any case, the percentages with SOC categories for any given month generally appear
to mirror both the Year 2 and 2-Year averages. The Year 2 findings indicate that of the jobs held by
participants: 13 percent of both treatment and control participants were in management, business,
science, or arts; 37 percent of treatment participants compared to 41 percent of controls held service
jobs; 37 percent of treatment participants compared to 34 percent of controls held sales and office
jobs; 2 percent of both treatment and control participants worked in construction or maintenance;
and 10 percent of treatment participants compared to 11 percent of controls worked in production
or transportation.
3
For each job reported, participants also reported the month and year the job started and stopped. Since a participant could start a job immediately
after study enrollment, the month in which a participant enrolled was the first calendar month for which employment outcome data were analyzed
(Month 1). Counting out 24 months from Month 1 yields a total of 25 calendar months for which MHTS employment outcomes were assessed.
Chapter 4: Outcomes
4-14
Table 4-7.
Study
month
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Baseline
freq
%
Year 2
M
%
Total
M
%
Types of occupations held by participants who reported working in a job for each
study-month by treatment and control group
Management,
business,
science, and
arts
occupations
Trmt
Ctl
10
11
14
17
18
20
26
21
30
17
31
16
34
26
40
22
34
19
42
26
31
26
35
26
41
25
39
21
38
21
50
26
38
24
35
19
39
27
35
21
33
17
43
22
39
25
39
21
40
18
Service
occupations
Trmt
Ctl
16
18
29
32
49
32
62
52
65
41
81
53
96
60
92
51
87
53
109
58
102
53
106
55
106
65
103
67
109
64
112
69
112
67
113
70
122
66
120
63
113
68
120
72
113
71
102
75
98
74
Sales and
office
occupations
Trmt
Ctl
12
20
21
28
44
40
67
46
70
39
83
49
95
63
93
60
85
59
95
64
93
56
94
52
112
57
107
54
102
49
115
58
112
55
120
53
120
59
101
55
96
54
120
62
117
55
118
56
111
58
Natural
resources,
construction,
and
maintenance
occupations
Trmt
Ctl
1
2
2
2
7
4
6
5
11
4
7
5
9
4
8
4
7
3
8
6
6
4
6
6
10
7
6
6
8
5
10
2
5
2
5
1
6
1
4
1
3
2
8
1
7
1
5
2
6
2
Production,
transportation,
and material
moving
occupations
Trmt
Ctl
9
9
10
18
15
21
25
27
21
22
22
23
33
26
24
25
26
21
29
24
33
24
32
21
35
21
29
17
30
13
37
17
34
16
38
19
39
24
35
22
26
21
29
18
27
17
24
13
23
12
71
15.8
74
13.6
119
26.6
170
31.3
182
40.6
202
37.1
12
2.7
18
3.3
64
14.3
80
14.7
39.2
13.1
22.1
13.3
111.0
37.1
68.5
41.1
111.6
37.3
55.8
33.5
6.4
2.1
2.5
1.5
31.2
10.4
17.7
10.6
33.5
13.6
21.1
14.0
90.5
36.8
56.4
37.4
89.4
36.3
50.9
33.8
6.2
2.5
3.2
2.1
26.6
10.8
19.2
12.7
Chapter 4: Outcomes
4-15
Tables 4-8 and 4-9 show the distributions of occupational types by diagnosis and by education.
These tables demonstrate the expected pattern of a higher percentage of managerial or business jobs
among participants with an affective disorder, as compared to participants with schizophrenia, and a
gradient of increasing proportions of participants achieving managerial jobs as a function of level of
education. In each case, there were correspondingly higher percentages of participants with
schizophrenia and with less education in service jobs.
Table 4-8.
Types of occupations held by participants who reported working in a job by
diagnosis
Treatment
Affective
disorder
%
M*
Management, business,
science, and arts
occupations
Service occupations
Sales and office occupations
Natural resources,
construction, and
maintenance occupations
Production, transportation,
and material moving
occupations
*The
Control
Schizophrenia
M*
%
Affective
disorder
M*
%
Schizophrenia
M*
%
26.2
15.2
7.3
10.0
18
16.9
3.2
7.3
53.1
70.5
30.7
40.8
36.9
19
50.6
26.1
37.9
38.2
35.5
35.9
18.8
12.3
42.5
27.8
4.3
2.5
2
27.8
1
0.9
2.2
5.0
18.7
10.8
7.8
10.7
11.5
10.8
7.7
17.4
reported mean is the average proportion across the 24-month study period.
Predicting employment. An important secondary question in the MHTS was whether it was
possible to predict which participants work, or under what circumstances they work. Three variables
were particularly useful for explorations of work and work behavior: obtained employment, total
months employed, and steady worker. These three variables offered the opportunity to explore
increasing levels of work behavior in participants who obtained jobs during the study. ―Obtained
employment‖ includes the largest number of participants in the study, counting any participant who
reported having a job in any month during the study. ―Total months employed‖ offers a different
dimension of obtained work, including the amount of work measured by the number of months in
which a participant reported having a job (up to 25). Third, the ―steady worker‖ variable defines a
special subgroup of participants in the MHTS. These individuals reported having a job for at least 10
months during the study. The goal of these analyses was to identify predictor variables using a
combination of logistic and zero-inflated negative binomial regression techniques to explain the
characteristics of working participants, events associated with working, or conditions that predict
work. The analyses involving categorical outcome variables (obtained employment and steady
Chapter 4: Outcomes
4-16
worker) used logistic regression. The analyses involving the numerical variable (total months
employed) used zero-inflated negative binomial regression.
Table 4-9.
Types of occupations held by participants who reported working in a job by education
Less than HS
M*
%
Treatment
Management, business,
science, and arts
occupations
Service occupations
Sales and office occupations
Natural resources,
construction, and
maintenance occupations
Production, transportation,
and material moving
occupations
Control
Management, business,
science, and arts
occupations
Service occupations
Sales and office occupations
Natural resources,
construction, and
maintenance occupations
Production, transportation,
and material moving
occupations
*The
HS graduate
M*
%
College
graduate
M*
%
Graduate
school
M*
%
0.6
1.9
13.9
10.0
12.4
20.4
6.6
42.7
11.8
9.8
39.5
32.9
57.0
50.3
40.9
36.2
17.0
25.2
27.9
41.4
4.3
3.7
27.8
23.8
1.5
5.1
4.2
3.0
0.5
0.9
0.0
0.0
6.1
20.5
13.7
9.9
5.7
9.4
0.9
6.0
0.7
4.2
7.5
8.9
10.8
24.8
2.3
32.9
8.4
5.0
53.4
32.1
33.8
27.2
40.0
32.2
12.5
16.0
28.8
36.6
1.9
2.3
27.9
32.9
0.3
2.2
2.5
3.0
0.3
0.8
0.0
0.0
1.2
7.8
13.6
16.0
3.9
9.0
0.5
6.7
reported mean is the average proportion across the 24-month study period.
A potentially large number of variables were available to assess employment across the entire study
population, including both treatment and control group participants. The criteria for selecting
predictor variables for the entire study population included an indication of importance in previous
research or variables of interest to SSA. Table 4-10 lists the demographic, clinical, and site-level
potential predictors included in the analysis for all participants. Demographic variables included age,
gender, education, and Supplemental Security Income (SSI) status. Clinical variables included mental
or physical health, diagnosis, emergency room visits at baseline, and hospital stays at baseline. Site
variables included urbanicity as measured by population density around the study site, and an
effective unemployment rate was computed by weighting the local unemployment rate in a given
year by the participant‘s length of participation in the study in that particular year. All of these
Chapter 4: Outcomes
4-17
variables were available on each participant enrolled in the study, including those in the treatment
and control groups.
Table 4-10.
Potential predictor variables included in employment-related regression models
Variable
Treatment dummy
Baseline mental health (SF-12)
Baseline physical health (SF-12)
Gender
Worked in last 2 years (pre-baseline)
Diagnosis
Age
Less than HS graduate vs. More than HS graduate
HS graduate vs. More than HS graduate
Unemployment rate
Months on rolls
Number of ER visits at baseline
Number of hospital stays at baseline
Population density
Not receiving SSI (SSDI only)
Ever had active ticket
SE engagement score
SMM engagement category 0 vs. 2
SMM engagement category 1 vs. 2
Received medication services
Received general medical care
Received substance abuse treatment
Received housing services
Received family counseling or legal assistance
Received social skills training
Received financial assistance training
Received case management
Average site fidelity score
Number of clients served at site
Treatment and control
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Treatment only
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
An even larger number of variables were available to assess employment among treatment group
participants only. The study collected a rich set of useful data on just treatment group participants,
most of which pertain to the treatment intervention, services received, or study site characteristics,
that were not particularly relevant to control group participants. Also shown in Table 4-10, the
additional potential predictors included in the regression models for treatment group participants
only included dummy variables indicating receipt of some other behavioral health or related service
(e.g., substance abuse treatment, case management, or social skills training) at some point during the
Chapter 4: Outcomes
4-18
24-month study period; variables measuring the extent to which a participant was engaged with the
IPS SE and systematic medication management (SMM) components of the intervention; size of
study site (measured by the number of clients served at the site); and the site fidelity scores.
The general strategy for all regressions (logistic and zero-inflated negative binomial) associated with
predictions of employment was to run unadjusted bivariable models using each predictor variable
with the outcome variable. Analyses using data on all participants included the treatment dummy
(i.e., in treatment group—yes vs. no) as a third variable in the unadjusted bivariable preliminary
models. Predictor variables with an estimate having a p-value of .25 or less were included in the final
multivariable model. Reporting in the tables that follow only include those models containing
variables with a p-value of .05 or less.
Table 4-11 presents three panels containing the final regression models for obtained employment,
total months employed, and steady worker. Overall the “obtained employment” model yielded an
acceptable fit (c=.709) and included 5 of the original 16 predictor variables (including treatment
dummy), two of which appear to be relatively important. Being in the treatment group, having better
physical health, working in the past two years at baseline, lower unemployment rates, and lower
number of months on the SSDI rolls were significant in the final model for “obtained employment.”
The MHTS treatment dummy had a positive and highly significant marginal effect estimate. The
results show that, on average, being in the treatment group increased the probability of obtaining
employment by 23 percent, ceteris paribus. Higher physical health status at baseline and having work
experience in the last 2 years (pre-baseline) significantly increased the probability of obtaining
employment. A unit increase in the local unemployment rate and number of months on SSDI rolls
significantly decreased the probability of obtaining employment.
The “steady worker” model repeated these same results with a few minor exceptions. In addition to
the same five variables (being in the treatment group, having better physical health, working in the
past two years at baseline, number of months on rolls and unemployment rates), being on the SSI
rolls and number of hospital stays at baseline figured into the “steady worker” model. Being in the
treatment group, on average, increased the probability of becoming a steady worker by 14 percent,
ceteris paribus. Being on the SSI rolls and a unit increase in the number of hospital stays at baseline
both reduced the probability of reporting steady employment.
Chapter 4: Outcomes
4-19
Table 4-11.
Estimates of marginal effects on obtained employment, steady worker, and number
of months employed (for all participants)
Variable
Estimate
p-value
Logistic regression
A. Obtained employment1
Treatment
Baseline physical health (SF-12)
Worked in last 2 years (pre-baseline)
Unemployment rate
Months on rolls
0.226
0.004
0.308
-0.019
-0.001
<0.001
<0.001
<0.001
0.017
0.005
B. Steady worker2
Treatment
Baseline physical health (SF-12)
Not receiving SSI (SSDI only)
Worked in last 2 years (pre-baseline)
Unemployment rate
Months on rolls
Number of hospital stays at baseline
0.139
0.002
-0.523
0.200
-0.019
-0.001
-0.048
<0.001
0.020
0.017
<0.001
0.001
0.001
<0.001
Zero-inflated negative binomial regression
C. Total months employed
Treatment
Baseline mental health (SF-12)
Baseline physical health (SF-12)
Number of hospital stays at baseline
Months on rolls
Worked in last 2 years (pre-baseline)
Ever had active ticket
Unemployment rate
2.573
0.029
0.042
-0.673
-0.008
3.836
0.849
-0.359
<0.001
0.022
0.002
0.002
<0.001
<0.001
0.054
0.001
1The
C-statistic for the Obtained Employment model is 0.709. The C-statistic ranges from.5 to 1 where C=1 for a perfect model and C=.5
for a model no better than random classification.
2The C-statistic for the Steady Worker model is 0.708. The C-statistic ranges from.5 to 1 where C=1 for a perfect model and C=.5 for a
model no better than random classification.
A zero inflated negative binomial regression model supported examination of the treatment
intervention impact on the count of ―total months employed‖ over the study period. The dependent
variable was total months employed, which carries a value up to 25 months with a large
concentration of ―zero‖ observations (i.e., never employed). Zero-inflated count models provide a
powerful way to model this type of situation. This model assumes that the data are a mixture of two
separate data generation processes: one generates only zeros and the other is a negative binomial
data-generating process. The result of a Bernoulli trial is used to determine which of the two
processes generates an observation.
Chapter 4: Outcomes
4-20
Table 4-11 presents marginal effect estimates for “total months employed,” for the backward
stepwise estimation including only covariates that are significant at.05 (i.e., the reduced model).
Marginal effect estimates in the reduced model indicated that the expected impact for each person
of participating in the MHTS (versus not participating in the MHTS), averaged across the entire
study sample (both control and treatment), was 2.57 additional months of employment. Baseline
mental health, baseline physical health, having prior work experience, and having an active ticket
were significant predictors with positive effects on the “total months employed” outcome. Number
of hospital stays, number of months on SSDI rolls, and the local unemployment rate were
significant with negative effects on the “total months employed” outcome.
Table 4-12 presents the results of similar regression estimates using treatment group only data.
These estimates are important because they offer a greater number of potentially important
predictor variables associated only with beneficiaries who participated in the treatment intervention.
For participants that obtained employment during the study, eight predictor variables were
significant in the final regression model. A unit increase in physical health score (at baseline),
engagement in SE, and population density all indicated a positive impact on the probability of
―obtained employment.‖ Participants who worked in the last 2 years at baseline also had a higher
likelihood of obtaining employment. A unit increase in local unemployment rate, number of months
on rolls, and number of hospital stays at baseline decreased the probability of obtaining
employment. Participants who received case management services also were less likely to report
employment.
The logistic regression results for the steady worker model were similar to the obtained employment
model with a few minor exceptions. Population density and physical health at baseline did not
appear to be statistically significant in the model for the steady worker. Instead, having an active
ticket (from SSA‘s Ticket to Work program) was significant and had a positive marginal effect on
the probability of being a steady worker. As with obtained employment, participants reporting that
they worked in the past two years (at baseline) were more likely to be steady workers than were
participants who did not report such work experience. A unit increase in the local unemployment
rate, number of months on rolls, and number of hospital stays all had negative marginal effects on
the probability of becoming a steady worker. Participants that received case management services
were also less likely to report steady employment.
Table 4-12 also presents the results of the zero-inflated negative binomial estimation for total
months employed, using data from the treatment group only. Seven predictor variables were
significant in the reduced regression model. Number of hospital stays, number of months spent on
Chapter 4: Outcomes
4-21
SSDI rolls, local unemployment rate, and receipt of case management all had a significant negative
effect; having a work experience in the last two years (at baseline), ever having an active ticket, and
level of engagement in SE had a significant positive effect on total months employed.
Table 4-12.
Estimates of marginal effects on obtained employment, steady worker, and number
of months employed (for treatment group participants only)
Variable
Logistic regression
A. Obtained employment1
Baseline physical health (SF-12)
SE engagement score
Worked in last 2 years (pre-baseline)
Unemployment rate
Months on rolls
Number of hospital stays at baseline
Receive case management
Population density
Estimate
p-value
0.004
0.018
0.221
-0.032
-0.001
-0.038
-0.100
0.029
0.006
<0.001
<0.001
0.003
<0.001
0.044
0.003
0.003
B. Steady worker2
Worked in last 2 years (pre-baseline)
Unemployment rate
Months on rolls
Number of hospital stays at baseline
Ever had active ticket
SE engagement score
Receive case management
0.192
-0.029
-0.001
-0.070
0.110
0.014
-0.094
<0.001
0.003
<0.001
0.002
0.017
<0.001
0.001
Zero-inflated negative binomial regression
C. Total months employed
Number of hospital stays at baseline
Months on rolls
Worked in last 2 years (pre-baseline)
Ever had active ticket
Unemployment rate
SE engagement score
Receive case management
-1.091
-0.013
3.675
1.770
-0.495
0.254
-1.401
0.001
<0.001
<0.001
0.010
0.001
<0.001
0.006
1The
C-statistic for the Obtained Employment model is 0.709. The C-statistic ranges from.5 to 1 where C=1 for a perfect model and C=.5
for a model no better than random classification.
2The C-statistic for the Steady Worker model is 0.708. The C-statistic ranges from.5 to 1 where C=1 for a perfect model and C=.5 for a
model no better than random classification.
Reasons for not working full-time. Reasons for not working full-time among all participants
reporting part-time work have been examined. During the final Followup interview, participants that
did not report working full-time during the previous quarter answered a followup question about
why. Table 4-13 shows the responses for those participants responding to the question. For both
Chapter 4: Outcomes
4-22
treatment (38.6 percent) and control group (50.6 percent) participants, the most frequent reason is
―Too sick to work full-time.‖ The difference between the treatment and control groups is statistically
significant (p-value = 0.009). Other most frequent reasons were ―Couldn’t find full time job‖ (15.7 and
15.9 percent for treatment and control group, respectively) and ―Don’t know‖ (14.5 and 13.2 percent
for treatment and control group, respectively).
Table 4-13.
Reasons for working part-time
Reasons
Couldn’t find full-time job
Too sick to work full-time
Don’t want to work more
Other demands on time (i.e., pets, child)
Make enough money working part-time
Didn’t want to lose SSDI or medical benefits
In school (full- or part-time)
Other reasons
Don’t know
Treatment
(N=332)
N
%
52
15.7
128
38.6
23
6.9
14
4.2
17
5.1
23
6.9
5
1.5
22
6.6
48
14.5
Control
(N=182)
N
%
29
15.9
92
50.6
9
5.0
0
0.0
2
1.1
12
6.6
1
0.6
13
7.1
24
13.2
p-value
0.936
0.009
0.374
0.005
0.021
0.886
0.334
0.824
0.691
Among treatment group participants, 6.9 percent of respondents stated that they did not want to
work more and the same percent reported they did not want to lose SSDI benefits or medical
benefits (provided by the Center for Medicare and Medicaid Services). While 6.6 percent of
treatment participants reported other reasons for working part-time, 1.5 percent of them were in
school on a full- or part-time basis. Among control group participants, 5.0 percent stated that they
did not want to work more (than they were currently working) and 6.6 percent reported they did not
want to lose benefits. The findings also show that 7.1 percent of treatment participants had other
reasons for working part-time and 0.6 percent reported going to school on a full- or part-time basis
as the reason for working part-time.
Reasons for not working. Many of the treatment group participants achieved some success in
working. As expected, some did not. It was just as important to find out as much as possible about
those who experienced little or no success in working. The literature on employment and mental
illness, reviewed earlier, emphasizes several factors, including uncontrolled symptoms of mental
illness, ongoing substance use, poor social skills, cognitive problems, and so forth.
From the beginning of the study, it was apparent that a small proportion of participants entered the
study with primary motivations other than to work. These participants understandably wanted better
Chapter 4: Outcomes
4-23
insurance, better health care, or some other benefit from the study, but they showed minimal or no
interest in working throughout their tenures in the study. Some never met with an SE specialist;
some engaged superficially but showed no interest in trying to work. Another problem encountered
from the beginning of the study was that a number of participants had serious physical health
problems that limited their ability to find a job and participate in employment. Apparently, physical
health problems were part of their disablement even though the primary listing on Social Security
records was mental illness. These two problems differed from the usual barriers to employment that
IPS specialists have confronted in previous studies. Of course, the two problems also reflected the
unique features of the MHTS. That is, beneficiaries entering this study often did not have adequate
health care insurance or were afraid of losing Medicare or Medicaid coverage, were not already
engaged in community mental health care, were older, and were more physically ill than the usual
participants in IPS because they were all SSDI beneficiaries. After randomization, the study had
relatively little contact with the control group, other than to conduct the quarterly interviews, and
little health data. Thus, these factors were assessed more fully for the treatment group. It was
assumed that these problems were equally distributed by randomization. Their similar SF-12 scores
at baseline on the physical scale partially validate this assumption.
The study has several ways to investigate lack of motivation, physical health, and other barriers in
the treatment group. Later chapters report on assessments of engagement in SE and of service
utilization. Reports of baseline predictors of employment appear earlier in this chapter. This section
considers barriers to competitive employment through ratings made by the clinical teams at the end
of each participant‘s participation in the study.
Barriers to competitive employment survey. The nurse care coordinator (NCC), along with the
SE specialist and other team members rated (as a group) the reasons for lack of employment among
those treatment group participants who worked little or not at all. The definition was less than one
month of employment during the 24-month study period. Using a fixed list of 14 common barriers from the
literature, the teams rated up to three reasons that a participant did not work, and prioritized the
factors as the first, second, and third most important, assuming that a combination of factors, rather
than just one factor, might have been important for many participants. Site staff completed 430
rating forms.
Table 4-14 shows a summary of the ratings on barriers to competitive employment on the 430
participants with completed rating forms. The number of completed ratings by site ranged from 5 to
41, in correspondence with the number of enrolled treatment participants at the site. Three factors
dominated the primary barriers category. For 34 percent of the group, poorly controlled symptoms
Chapter 4: Outcomes
4-24
of mental illness were the primary barrier; these participants had unremitting or recurrent and severe
symptoms despite SMM (if they were even engaged with SMM). For 26 percent of the group, failure
to engage in SE was primary; these clients showed little or no interest in meeting with a SE specialist
from the beginning. For another 16 percent, poorly controlled physical health problems were
primary; these participants were generally more disabled by their severe emphysema, diabetes, or
other physical illnesses than by their mental illnesses. As primary barriers, one of these three
problems affected over 75 percent of the participants who were unsuccessful in working. No other
factor was rated as primary for even 10 percent of the group.
Table 4-14.
Barriers to competitive employment for non-working treatment group participants1
Priority 1
Barrier/Factor2
Failure to engage in SE
Priority 2
Priority 3
Priority
1, 2, or 3
n
111
%
25.8
n
52
%
12.1
n
26
%
6.1
n
189
%
44.0
Disengagement in SE
24
5.6
28
6.5
16
3.7
68
15.8
Physical health problems not
controlled
Substance abuse/dependence not
well controlled
Symptoms of mental illness not
well controlled
Cognitive problems
70
16.3
48
11.2
23
5.4
141
32.8
37
8.6
31
7.2
11
2.6
79
18.4
144
33.5
75
17.4
17
4.0
236
54.9
5
1.2
15
3.5
11
2.6
31
7.2
10
2.3
48
11.2
23
5.4
81
18.8
Family problems
Lack of needed services (e.g., case
management)
Transportation
3
0.7
10
2.3
11
2.6
24
5.6
2
0.5
7
1.6
7
1.6
16
3.7
Lack of work skills
6
1.4
5
1.2
6
1.4
17
4.0
Lack of social skills
4
0.9
13
3.0
18
4.2
35
8.1
Criminal justice system problems
3
0.7
9
2.1
9
2.1
21
4.9
Housing problems
1
0.2
6
1.4
13
3.0
20
4.7
Behavior problems
11
2.6
17
4.0
20
4.7
48
11.2
1Includes
participants who did not experience employment success, defined as no employment or only brief employment of less than
one month. Rating forms were completed for 430 participants.
2Up to 3 barriers were rated as a priority for each participant. For some participants only 1 or 2 barriers were rated as priorities.
As a percentage of the total treatment group (N=1,004), these figures allow us to estimate the
influence of primary barriers for participants who did not experience employment success. For
example, 111 of the 1,004 treatment group participants, or 11 percent, never engaged in SE. These
individuals may have entered the study for reasons other than pursuing employment. For another
144 (14%) of these 1,004 participants the investigative team judged them to have been unsuccessful
in work because their symptoms of mental illness were poorly controlled. In addition, another 70
Chapter 4: Outcomes
4-25
(7%) were unable to work primarily because of physical health problems. These percentages are
likely underestimates because the percentages only include participants who worked less than one
month during the two years of the study and had a completed rating on barriers to competitive
employment.
Further inspection of Table 4-14 reveals that for many participants, multiple barriers were pertinent.
No single factors were highly prevalent as secondary or tertiary barriers, but the sum of barriers
(primary, secondary, and tertiary) indicated that several factors were often involved. Poorly
controlled symptoms of mental illness affected 55 percent of the group; 44 percent had difficulties
with engagement; and 33 percent had poorly controlled physical health problems. In relation to the
entire treatment group, poorly controlled symptoms of mental illness impeded work for 23 percent
of participants, lack of engagement for 18 percent, and physical health problems for 13 percent. Of
course, the total proportions with these problems were probably much higher because many
individuals with similar problems probably worked. In addition to these three major barriers, family
problems, poorly controlled substance use disorders, disengagement from SE services, and behavior
problems affected greater than 10 percent of the group.
Factors with extremely low ratings (less than 10%) indicate that treatment group participants in the
study did not apparently encounter them. These factors included cognitive problems, lack of
services, transportation difficulties, lack of work skills, lack of social skills, criminal justice system
problems, and housing problems.
Understanding the impact of work on Social Security benefits. Integrated service provisions at
each study site also included some form of benefit counseling for treatment group participants. In
order to assess whether there was any significant difference over time in understanding and gaining
knowledge of the SSDI program benefits, both treatment and control group participants agreed or
disagreed in interviews with several statements about their Social Security benefits. Table 4-15 shows
the percent of participants who ―agreed‖ with each statement about their Social Security benefits.
The baseline and final followup interviews included the same items.
The treatment group had a larger decrease in the percent of participants agreeing with the statement:
―As soon as people start working they stop getting their benefit checks.‖ For the treatment group, participants
who agreed with this statement went from 16 percent at baseline to 9 percent at followup. The
control group had a decrease from 16 percent to 14 percent (p-value = 0.019). Comparisons
between the changes observed for the treatment and control groups yield significant increases in the
percent of participants who agree with the following statements: ―I can make more money at a job and
still collect my benefit checks‖ (77% to 81% vs. 78% to 69%, p-value < 0.001); ―As soon as people start
Chapter 4: Outcomes
4-26
working they lose their medical coverage‖ (15% to 12% vs. 14% to 18%, p-value = 0.004); ―Unless a job offers
coverage of mental health and prescriptions, I can’t afford to take it‖ (42% to 39% vs. 40% to 44%, p-value =
0.007). The percent of treatment participants who agree with the statement ―If I go to work, get off
benefits and get sick right away, I’ll have a hard time getting back on benefits‖ went down from 50 percent to
40 percent while there was very minimal change in the control group (p-value < 0.001). All these
significant changes in beliefs in the direction favoring improved knowledge of the treatment group
indicate that the benefit counseling was effective in explaining availability of work trial benefits and
other options to the treatment participants.
Table 4-15.
Agreement with statements about Social Security Disability benefits programs by
treatment and control group
Statements
Treatment (N=1,004)
Baseline
Exit
d
Control (N=1,051)
Baseline
Exit
d
As soon as people start working they stop getting their benefit checks.
n
163
90
167
%
16.4
-7.6
7.6
16.4
142
13.7
p-value
-2.7
I can make more money just collecting my benefit check than I can if I go to work while on benefits.
n
112
115
107
129
%
11.1
11.5
0.4
10.1
12.2
2.1
0.019
0.315
I can make money at a job and still collect my benefit checks.
n
774
815
%
77.2
81.3
4.1
826
78.4
730
69.1
-9.3
<0.001
As soon as people start working they lose their medical coverage.
n
150
125
%
15.2
12.5
-2.7
149
14.4
183
17.7
3.3
0.004
4.1
0.007
Unless a job offers coverage of mental health and prescriptions, I can’t afford to take it.
n
415
384
416
461
%
41.6
38.5
-3.1
39.8
43.9
If I go to work, get off benefits and get sick right away, I’ll have a hard time getting back on benefits.
n
502
398
502
498
%
50.5
39.7
-10.8
48.0
47.5
-0.5
I can’t afford to get training to help me get a better job.
n
546
432
%
54.4
43.3
-11.1
567
54.2
456
43.6
<0.001
-10.6
0.695
If I knew that I wouldn’t lose all of my benefits, I would try to get a job or get a better job.
500
734
513
n
732
%
73.1
50.2
-22.9
69.8
48.8
-21.0
0.413
NOTE: Weighted percents may not be consistent with unweighted counts.
Chapter 4: Outcomes
4-27
There were no statistically significant differences between the treatment and control groups in terms
of a percent change in agreement with the statements ―I can’t afford to get training to help me get a better
job‖ and ―If I knew that I wouldn’t lose all of my benefits, I would try to get a job or get a better job.‖ In both
cases, the percentage of participants agreeing with the statement dropped in favor of improved
knowledge, but the drop was consistent across both treatment and control group, and therefore, not
significant.
Earnings and Income
Results on earnings rely on data from two different questionnaire sources: self-reports on earnings in
the past month (collected at each of the 8 post-baseline interviews) and self-reports on wages per
hour and average hours per week in each job in the past three months (collected only at the final
Followup interview).
Past month’s earnings. Past month‘s earnings were averaged for each participant over all 8 postbaseline interviews. These averages are shown in Table 4-16. The mean monthly earnings for the
treatment group were $148.16 versus $97.41 for the control group, a difference that is both
statistically and substantively significant (p-value < 0.001). A large portion of this difference in
means was due to the much higher percentage of the treatment group participants reporting positive
earnings in one or more of the eight interviews than did control group participants (58.67% vs.
42.72%, p-value < 0.001). However, if the comparison is restricted to only those participants
reporting any earnings, the difference in means ($251.12 vs. $227.93) was also significant (p-value <
0.001). The corresponding difference in median earnings ($165.37 vs. $93.98) was even larger and, as
is usually the case with earnings data and as is implied by the extent to which the means exceed the
median, the distribution of mean earnings was highly skewed to the right. Relatively large differences
(with higher figures for the treatment group) were also observed at the 25th and 75th percentiles of
the distributions of average monthly earnings (for those who reported any positive average).
Past three months’ earnings. The data on earnings in the prior three months from the final
Followup interview (Table 4-16) showed significantly higher treatment group mean earnings
($858.60 vs. $478.87, p-value < 0.001) and significantly higher treatment group percentages for
having any earnings (33.8% vs. 17.5%, p-value < 0.001).
Chapter 4: Outcomes
4-28
Table 4-16.
Past month’s earnings (averaged over 8-post baseline interviews), past three
months’ earnings (at study exit) and earnings above SGA (averaged over 8-post
baseline interviews) by treatment and control group
Unconditional
(Includes zero earnings)
Variable
Treatment
Control
d
Past month’s earnings (averaged)
N (n)
1,004
1,051
Mean
148.16
97.41
50.75
SD
257.84
283.34
Median
23.79
0.00
23.79
% with earnings >0
Past 3 months’ earnings (at study exit)
N (n)
1,004
1,051
Mean
858.60
478.87
SD
1,752.36 1,698.82
Median
0.00
0.00
% with earnings >0
379.73
p-value
<0.001
<0.001
0.00
Earnings above SGA
n
%
Conditional
(Excludes zero earnings)
Treatment
Control
d
589
251.12
294.32
165.37
59.00
449
227.93
397.97
93.98
42.73
340
2,538.16
2,203.52
1,935.59
33.83
184
2,739.25
3,233.26
1,910.20
17.48
49
8.18
40
8.76
p-value
23.19
<0.001
71.39
16.27
<0.001
-201.29
0.607
25.39
16.35
<0.001
-0.58
0.7394
NOTE: Weighted percents may not be consistent with unweighted counts.
Substantial gainful activity (SGA). Table 4-16 also presents treatment group versus control group
differences in the percent of participants reporting past month‘s earnings (averaged across all eight
post-baseline interviews) was above the current SGA level of $1,000. The differences were generally
fairly small and not significant (p-value = 0.7294). For all participants, the relevant percentages were
8.2 percent (treatment) and 8.8 percent (control).
Investigators and policy makers have observed that only very small numbers (~4%) of SSA disability
participants leave the rolls by working at jobs with earnings above SGA (Stapleton, 2010). This
phenomenon is true in spite of successful SE interventions that assist participants in returning to
competitive employment. Few SE participants work above SGA and leave the disability rolls. Some
policy analysts have observed that even when participants return to work, they seem to work
intentionally below the SGA level to avoid losing their benefits. There is speculation and anecdotal
evidence that this occurs because participants fear losing their cash payments and their access to
health insurance if they leave the rolls, even if they leave for better paying jobs. This could occur
because of a lack of awareness of trial work programs and other ways to return to work and not lose
benefits or because participants are afraid that if they relapse from their underlying impairments or
they lose their current work arrangement, they will have difficulty returning to benefits in the
Chapter 4: Outcomes
4-29
disability program. It took many workers a long time to get on the program, having declared
themselves unable to work at any job or perform SGA, and they fear losing those benefits. The
phenomenon of working just below the SGA level is called ―parking‖ (Schimmel, Stapleton, & Song,
2010; Porter, Smith, Payette, Tremblay, Burt, 2009), and the MHTS tested the distribution of
earnings to determine if MHTS participants exhibited earnings distribution consistent with the
―parking‖ phenomenon. It should be noted that a recent GAO report on the effect of SGA levels
on work behavior indicated that very few participants exhibited the ―parking‖ phenomenon.
A distribution for each of the 8 quarters of employment reported by participants in the MHTS—
both participants in the treatment and control conditions—was examined to see what proportion of
individuals had earnings that fell into each of three categories of earnings: above SGA, between
SGA and 75 percent of SGA, and below 75 percent of SGA. Ideally, the MHTS would have been
successful if many participants were able to work steadily above SGA and leave the disability
program entirely—earning higher wages and salaries and obtaining health insurance in the
workplace. First, workers had earnings all across the earnings distribution but mostly below SGA
indicating that return to work was difficult for MHTS participants and that SGA might be a natural
barrier for work capacity for individuals. Second, workers worked many quarters in the range of 75
percent of SGA but below SGA because they were capable of considerable earnings from work but
deliberately chose to work below SGA to avoid loss of cash benefits (―parking‖).
Table 4-17 shows the number of MHTS participants who have earnings in three different ranges of
earnings. Note that some workers will report earnings in all three distributions as the observation is
persons by quarter and not an unduplicated count of workers between the three tables reporting on
three different ranges of earnings.
Across eight quarters, only 13.5 percent of participants in the treatment group reported earnings
above SGA in the last month of one or more quarters, and only 5.6 percent did so for two or more
quarters. The comparable figures for the control group were 8.2 percent and 3.4 percent,
respectively. The vast majority of those who did report a quarter of work above SGA report
earnings at that level in only one or two quarters. Only ten workers in both groups combined
worked six or more quarters above SGA.
Among the participants that reported work between 75 percent and 99 percent of SGA, 15.9 percent
of those in the treatment group reported at least one quarter of work and only 6.4 percent did so for
two or more quarters. The comparable figures for the control group were 7.9 percent and 2.3
percent, respectively. Again, as observed for work above SGA, the vast majority of those reporting
work in this range do so for only one or two quarters. In fact, only five treatment group members
Chapter 4: Outcomes
4-30
report six or more quarters in this range, and none in the control group report six quarters or more
above 75 percent of SGA but below SGA itself. An assessment of earnings in the latter four quarters
of the study did not yield anything different from above results (not shown).
Table 4-17.
0
1
2
3
4
5
6
7
8
Q1-Q8
Number of quarters with past month’s earnings above, near, or below SGA by
treatment and control group
Greater than or
equal to SGA
Treatment
Control
N
%
N
%
868 86.5
963 91.6
80
7.9
50
4.8
29
2.8
13
1.2
7
0.7
9
0.8
7
0.7
6
0.6
9
0.9
4
0.4
0
0.0
2
0.2
2
0.2
3
0.3
2
0.2
1
0.1
136 13.5
88
8.2
Between 75% and
99% SGA
Treatment
Control
N
%
N
%
844 84.1
968 92.1
97
9.7
59
5.6
38
3.8
16
1.5
16
1.6
2
0.2
3
0.3
5
0.5
1
0.1
1
0.1
3
0.3
0
0.0
1
0.1
0
0.0
1
0.1
0
0.0
160 16.1
83
7.9
Below 75% SGA
Treatment
Control
N
%
N
%
447 44.5
630 59.9
168 16.7
170 16.2
116 11.6
97
9.2
90
9.0
47
4.5
65
6.5
28
2.7
55
5.5
34
3.2
32
3.2
21
2.0
19
1.9
17
1.6
12
1.2
7
0.7
557 55.8
421 39.9
NOTE: Weighted percents may not be consistent with unweighted counts.
Although earnings were significantly higher for participants in the treatment group compared with
those in the control group, few people were earning greater than SGA. Furthermore, there was little
work yielding earnings in the range of 75 percent to 99 percent of SGA. These data suggest that
there is little evidence of ―parking‖ of earnings just below the SGA level to avoid loss of benefits by
participants returning to work who ―could‖ work more but do not do so to avoid loss of benefits.
MHTS participants may be aware of the SGA level of earnings and set target earnings below this
level, but it does not appear to result in a sustained pattern of working to receive earnings just below
SGA.
Income from sources other than earnings. Each post-baseline interview collected data for a
number of other sources of income received in the past month. These sources include SSDI, SSI,
SSA retirement or survivors benefits, Veterans Affairs (VA) benefits, other public welfare, food
stamps or Temporary Assistance for Needy Families (TANF), vocational programs, and
unemployment compensation. For many of these sources, the percentage of participants who
participated was very low (5% or less) reporting receipt of any income from that source. The
exceptions were SSDI benefits and SSI benefits, as well as income from family members, food
stamps or TANF, and a residual ―other income‖ category.
Chapter 4: Outcomes
4-31
Results in Table 4-18 indicate that mean SSDI income participants received was almost identical for
treatment ($855.49) and controls ($853.00). The SSI income participants received in the past month
also was not significantly different between the treatment and control groups. The analysis included
an examination of the past month‘s income from other public programs (averaged over the 8 postbaseline interviews) by treatment and control group.
Table 4-18.
Past month’s SSDI and SSI income (averaged over 8 post-baseline interviews) by
treatment and control group
Unconditional
(Includes zero income)
Treatment
Control
d
Variable
SSDI
N (n)
1,004
Mean
855.49
SD
340.30
Median
787.48
% with income >0
1,051
853.00
337.75
784.22
p-value
2.49
0.695
3.26
Conditional
(Excludes zero income)
Treatment
Control
D
1,003
856.34
339.40
787.54
59.00
1,050
853.77
336.97
784.60
99.90
241
93.41
89.52
59.54
24.50
259
105.37
94.71
79.08
24.61
p-value
2.57
0.692
2.94
-40.90
0.945
-11.96
0.175
-19.54
-0.11
0.952
SSI
N (n)
1,004
Mean
22.48
SD
60.65
Median
0.00
% with income >0
1,051
25.94
65.40
0.00
-3.46
0.854
-4.75
NOTE: Weighted percents may not be consistent with unweighted counts.
Results in Table 4-19 indicate that the mean SSA retirement or survivors benefits were similar for
treatment ($117.01) versus control ($115.74) groups. The results were similar for the past month‘s
income reported from other public welfare ($79.99 vs. $66.21; p-value = 0.906) and food stamps or
TANF ($74.26 vs. $78.96; p-value = 0.269). Mean VA benefits in the past month were higher among
the treatment group than the control group, but again, the difference was not statistically significant
($689.81 vs. $351.73; p-value = 0.102). Similarly, income from vocational programs ($58.66 vs.
85.44; p-value = 0.173) and unemployment compensation ($133.25 vs. $169.39; p-value = 0.773)
seemed to be lower among participants in the treatment group; however, neither of these differences
was statistically significant. There were more participants in the treatment group that reported nonzero income from vocational programs (6.03% vs. 2.07%; p-value < 0.001) and unemployment
compensation (3.62% vs. 2.28%; p-value = 0.071). These results indicate that treatment and control
groups were very similar in reporting past month‘s income from other public programs.
Chapter 4: Outcomes
4-32
Table 4-19.
Past month’s income from other public programs (averaged over the 8 postbaseline interviews) by treatment and control group
Unconditional
(Includes zero income)
Variable
Treatment Control
d
SSA retirement or survivors benefits
N (n)
1,004
1,051
Mean
4.01
5.44
-1.43
SD
39.16
33.44
Median
0.00
0.00
0.00
% with income >0
p-value
0.139
Conditional
(Excludes zero income)
Treatment Control
d
34
117.01
180.53
40.42
3.42
50
115.74
109.66
84.36
4.70
p-value
1.27
0.211
-43.94
-1.28
0.146
338.08
0.102
40.09
0.51
0.532
13.78
0.906
2.00
0.06
0.962
-4.70
0.269
-5.86
-1.76
0.427
-26.78
0.173
-9.86
3.96
<0.001
-36.14
0.773
-26.46
1.34
0.071
VA benefits
N (n)
Mean
SD
Median
% with income >0
1,004
1,051
26.05
216.02
0.00
11.52
118.38
0.00
Other public welfare
N (n)
Mean
SD
Median
% with income >0
1,004
7.54
41.48
0.00
Food stamps or TANF
N (n)
Mean
SD
Median
% with income >0
1,004
36.79
65.76
0.00
Vocational program
N (n)
Mean
SD
Median
% with income >0
1,004
3.54
33.62
0.00
Unemployment compensation
N (n)
1,004
Mean
4.82
SD
36.80
Median
0.00
% with income >0
1,051
6.20
31.99
0.00
1,051
40.51
72.39
1.18
1,051
1.77
22.59
0.00
1,051
3.85
39.60
0.00
14.53
0.00
1.34
0.959
0.00
-3.72
0.290
-1.18
1.77
<0.001
0.00
0.97
0.00
NOTE: Weighted percents may not be consistent with unweighted counts.
Chapter 4: Outcomes
0.522
4-33
0.072
39
35
689.81
917.00
141.15
3.78
351.73
567.37
101.06
3.27
94
79.99
110.56
36.65
9.42
99
66.21
83.96
34.65
9.36
485
74.26
75.88
47.83
49.54
539
78.96
84.88
53.69
51.30
62
58.66
127.45
12.79
6.03
22
85.44
135.06
22.65
2.07
37
133.25
148.74
64.72
3.62
24
169.39
207.68
91.18
2.28
Individual and household income. An examination of total income from all sources for the
individual (averaged across the 8 post-baseline interviews) and total household income (at study exit)
is presented below. Given the observed differences in past month‘s earnings noted earlier, it is
relevant to see whether the increased earnings for the treatment group translated into higher total
income or were offset by reductions in income from other sources, and particularly from reductions
in SSI or SSDI benefit payments.
Table 4-20 shows the total monthly individual income averaged over the study and the total monthly
household income at exit for participants in the treatment and control groups. The analysis reveals a
highly significant difference in the mean value of average monthly overall individual income for all
participants ($1180.52 treatment vs. $1120.96 control; p-value < 0.001). Presumably, this larger mean
income observed for treatment group participants is due almost entirely to the larger mean earnings
for these same participants relative to controls. Results for comparisons of total household income
reported at the final Followup interview between treatment and control group participants are also
presented in the table below. The treatment group reported higher monthly total household income
($1,678.30 vs. $1,661.56; p-value = 0.063), though the differential was somewhat smaller than that
observed for individual earnings or for total individual income.
Table 4-20.
Past month’s total individual income (averaged over 8 post-baseline interviews) and
past month’s total household income (at study exit) by treatment and control group
Unconditional
(Includes zero income)
Variable
Treatment
Control
d
Past month’s total individual income
N (n)
1,004
1,051
Mean
1,180.52 1,120.96
59.56
SD
495.14
525.64
Median
1,072.84
987.05
85.79
% with earnings >0
Past month’s total household income
N (n)
1,004
1,051
Mean
1,678.30 1,661.56
SD
1,180.98 1,248.01
Median
1,296.63 1,199.85
% with earnings >0
16.74
p-value
<0.001
0.063
96.78
NOTE: Weighted percents may not be consistent with unweighted counts.
†Not applicable. Chi-square not calculated.
Chapter 4: Outcomes
4-34
Conditional
(Excludes zero income)
Treatment
Control
d
1,004
1,180.52
495.14
1,072.84
100.00
1,051
1,120.96
525.64
987.05
100.00
997
1,689.90
1,176.95
1,298.11
99.31
1,048
1,666.51
1,246.50
1,201.46
99.70
p-value
59.56
<0.001
85.79
0.00
†
23.39
0.043
96.65
0.213
Trends in earnings differentials. The clearest treatment effects were on earnings and on total
income figures that included earnings. However, further exploration assessed the time pattern of
emergence of these earnings effects. Given the emphasis in IPS SE on rapid job placement, one
might expect that the treatment effect on earnings would be greater earlier in the 24-month followup
period and less pronounced later in the period as control group participants found work. To test this
hypothesis, a regression model was used to compute the linear time trend of past month‘s earnings
for each participant over the eight post-baseline interviews. Averages for the individual time trend
coefficients were computed for treatment and control group participants, and then tested for
differences in the mean values of these time trend coefficients between the two groups.
The results in Table 4-21 indicate that the mean quarterly increase in the past month‘s earnings
figure was $13.20 for the treatment group versus $4.25 for the control group; this difference was
highly significant (p-value < 0.001). In summary, the results suggest that the effect of the treatment
on earnings in fact increased over the 24-month followup study period.
Table 4-21.
Past month’s earnings regression slope (averaged over 8 post-baseline interviews)
by treatment and control group
Unconditional
(Includes zero earnings)
Variable
Treatment Control
d
p-value
Past month’s earnings
N (n)
1,004
1,051
Mean
13.20
4.25
8.95
<0.001
SD
63.52
55.86
Median
-0.04
-0.04
0.00
Conditional
(Excludes zero earnings)
Treatment Control
d
p-value
589
22.38
81.29
9.54
449
9.95
85.17
-0.81
12.43
<0.001
10.35
NOTE: Weighted percents may not be consistent with unweighted counts.
Earnings and income by subgroups. For many of the earnings and income outcome measures,
Table 4-22 presents a summary of the results of significance tests that were conducted to determine
whether differences between treatment and control group participants were statistically significant
for the following subgroups: age, gender, diagnosis, and education. These subgroup comparisons are
presented for the same three summary statistics analyzed for the overall comparisons—
unconditional means (i.e., includes reports of zeroes), conditional means (i.e., excludes reports of
zeroes), and percent reporting non-zero earnings and income amounts. Summary descriptive
statistics and the results of the significance tests are presented in Appendix 4C for all measures and
all subgroups shown in the table below.
Chapter 4: Outcomes
4-35
Table 4-22.
Statistically significant earnings and income subgroup comparisons for age, gender,
diagnosis, and education for unconditional mean comparisons, conditional mean
comparisons, and percent reporting non-zero amounts comparisons
Variables
Unconditional mean comparisons
Past month’s earnings (averaged)
Past 3 months’ earnings (at study
exit)
Past month’s SSDI
Past month’s SSI
Past month’s total individual income
Past month’s total household income
Past month’s earnings regression
slope
Age
18 to
35
35+
*
Conditional mean comparisons
Past month’s earnings (averaged)
Past 3 months’ earnings (at study
exit)
Past month’s SSDI
Past month’s SSI
Past month’s total individual income
Past month’s total household income
Past month’s earnings regression
slope
% Reporting non-zero amounts comparisons
Past month’s earnings (averaged)
*
Past 3 months’ earnings (at study
exit)
Past month’s SSDI
Past month’s SSI
Past month’s total individual income
Past month’s total household income
Past month’s earnings regression
slope
*
Gender
Diagnosis
Education
M
F
A
S
< HS
HS grad
> HS
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
p-value < 0.05
Analyses of treatment group versus control group differences in average past month‘s earnings were
also reported by age group (18–34 vs. 35+), by gender, by diagnosis (affective disorder vs.
schizophrenia), and by education level (less than high school, high school graduate, and more than
high school) (see Appendix 4C). The pattern of results is generally very similar to that described for
the overall group; however, there were some minor differences. Rates of participants reporting any
Chapter 4: Outcomes
4-36
positive earnings, which were higher for younger participants, were still significantly higher for the
treatment group, but the difference in means among those with positive earnings was not significant.
Since the magnitude of this difference in means was even larger for younger participants, the lack of
statistical significance was largely a result of very small numbers.
While the percent reporting non-zero past month‘s earnings was greater for females in the treatment
group, the conditional mean of past month‘s earnings among females was not significant.
Conditional mean past month‘s earnings were also not significantly different between the treatment
group and the control group for several other subgroups: participants 18 to 35, participants with
schizophrenia, participants with a high school diploma, and participants with less than a high school
education. In the case of participants with schizophrenia, the treatment versus control difference in
the conditional mean was positive but small; in the case of participants with less than a high school
education, this difference was large but the numbers of participants were small. As noted above, the
pattern of results for all other subgroups was similar to that for the entire study.
With regard to differences in the percentage of participants with average earnings above SGA for
subgroups, similar to the comparison for the overall group, no subgroup comparisons were
statistically significant, largely due to the very small sample size for this measure. Comparisons of
past three months‘ earnings (at study exit) revealed statistically significant differences between
treatment and control unconditional mean earnings, as well as significantly higher treatment group
percentages for having any earnings, for all subgroups except the small number of participants who
were under age 35 (for whom no significant treatment vs. control differences exist). On the other
hand, comparisons between conditional means showed little difference between treatment and
control for any subgroups of participants.
Analysis of total individual income revealed differences in the same direction that were also generally
significant for almost all respondent subgroups; the only exception was the non-significant
difference for participants under age 35. The only subgroups that showed significantly greater
household income for the treatment group were participants under age 35, males and females,
participants with schizophrenia, and those participants with more than a high school education.
Estimates of MHTS Impacts on Earnings and Income
In the previous section, the descriptive data presented used the imputed data for ease of
understanding the results. In this section, the analysis focuses on the results from multivariate
Chapter 4: Outcomes
4-37
analyses of MHTS effects earnings and income. Heteroscedasticity is a particular concern that
complicates the multivariate analyses due to unequal variances between groups. For example, the
large number of participants with zero earnings creates skewed distributions in some of the
comparisons. Because of these issues with heteroscedasticity (unequal variances), the multivariate
analysis used the unimputed data.
Data on dependent variables. The dependent variable measures originate from questions asked in
each of the eight followup interviews concerning earnings and income from other sources in the
immediate past 30 days prior to the interview date. The analysis includes estimates of MHTS impacts
for three different followup periods: quarters one through eight (i.e., two years of MHTS), quarters
one through four (i.e., first year of the MHTS), and quarters five through eight (i.e., second year of
the MHTS).
Earnings variables. The analyses used two different sets of dependent variables for earnings. One
definition uses the average past month‘s ―earned income or money from all paid employment,
including tips or commissions.‖ The second definition uses the same average earned income figure
plus the average of the past month‘s informal earnings; the latter figure based on responses
pertaining to ―money received by doing odd jobs such as babysitting or yard work, helping in a
business, or doing work ‗under the table‘.‖ For brevity, these two types of earnings variables are
referred to as ―formal‖ earnings and ―formal plus informal‖ earnings.
Descriptive statistics on these dependent variables pertaining to earnings, and their corresponding
baseline values, are shown in Table 4-23. Comparing the overall baseline values with the values for
the first four quarterly followup interviews (year 1) and the second four quarterly followup
interviews (year 2), a clear upward trend is discernable. For formal earnings, the upward trend was
from a baseline mean of $18.69 to average monthly figures of $103.25 and $139.73 for year 1 and
year 2 respectively. Analogous figures for formal plus informal earnings are $30.76, $110.67, and
$146.82, respectively. Comparisons between treatment vs. control groups show somewhat higher
baseline earnings figures for the control group, a strong upward trend in earnings for both groups
(especially through the first year of followup), substantially higher earnings for the treatment group
in followup quarters 1-4, and an even greater excess of treatment group earnings versus controls in
year 2.
Chapter 4: Outcomes
4-38
Income variables. Five different income measures are examined below:
1.
Disability income from Social Security (SSDI + SSI);
2.
All public sector income (SSDI + SSI +Social Security retirement or survivors benefits
+ VA + general assistance + food stamps or TANF + vocational programs +
unemployment compensation);
3.
Private non-earned income;
4.
Total non-earned personal income (public + private); and
5.
Total personal income (earned + unearned).
Table 4-23 presents the descriptive statistics on average values of each of these five variables along
with corresponding baseline values over the 24-month followup period.
Average disability income from social security (SSDI + SSI) was almost the same for both treatment
and control groups in the baseline as well as the followup period, and was only very slightly higher in
the followup period. A similar pattern shows for all public sector income supports, though the
baseline value for the control group was slightly lower than for the treatment group, thus the
increase in this figure for the followup period was slightly greater for the control group. Private
sector non-earned income was small relative to the income from social security (SSDI + SSI figure)
and the overall public sector figure, but it is interesting to note that the baseline value for private
sector non-earned income was modestly higher for the treatment group. The followup figures were
almost identical for the two groups, with the treatment group showing a very small decline from
baseline to followup while the control group showed a modest increase.
Total non-earned income was about 4 percent higher in the baseline for the treatment group
($1,035) than the controls ($996). Followup values for total non-earned income were almost
identical between the two groups, because of a very small increase from baseline for the treatment
group and a slightly larger increase for the controls. Finally, note that total personal income was only
about 3 percent higher at baseline for the treatment group versus the controls ($1,063 vs. $1,031),
while the increase from baseline to followup was somewhat greater for the treatment group,
resulting in a differential of about 5 percent between the treatment and controls at followup ($1,194
vs. $1,130). This differential likely occurred because of the differential growth of earnings for the
treatment group.
Chapter 4: Outcomes
4-39
Table 4-23.
Descriptive statistics for dependent measures included in multivariate analyses of
the MHTS impact on earnings and income by treatment and control group
Variables
Dependent variables
Avg. formal earnings
(Q1–Q8)
Avg. formal earnings
(Q1–Q4)
Avg. formal earnings
(Q5–Q8)
Avg. formal + informal
earnings (Q1–Q8)
Avg. formal + informal
earnings (Q1–Q4)
Avg. formal + informal
earnings (Q5–Q8)
Avg. SSDI + SSI income
(Q1-Q8)
Avg. total public income
support (Q1–Q8)
Avg. private non-earned
income (Q1–Q8)
Avg. total non-earned
income (Q1–Q8)
Avg. total personal income
N
All observations
M
SD
N
Treatment only
M
SD
N
Control only
M
SD
2,079
120
281
1,015
146
272
1,064
94
288
2,072
103
282
1,014
119
265
1,058
87
298
1,999
140
351
971
176
355
1,028
103
343
2,079
127
285
1,015
153
274
1,064
101
292
2,072
111
286
1,014
127
268
1,058
94
303
1,999
147
354
971
183
357
1,028
110
347
2,075
885
345
1,011
884
342
1,064
886
348
2,075
961
391
1,011
967
412
1,064
956
369
2,078
72
314
1,015
72
338
1,063
73
288
2,071
2,071
1,035
1,162
528
602
1,010
1,010
1,039
1,194
551
608
1,061
1,061
1,030
1,130
504
595
0.50
0.50
2,074
19
107
1,013
16
96
1,061
21
117
2,068
31
147
1,009
27
117
1,059
35
172
2,011
1,047
548
984
1,063
607
1,027
1,032
482
2,028
947
439
993
962
499
1,035
932
368
2,068
69
264
1,009
75
302
1,059
63
218
2,018
1,016
518
988
1,035
580
1,030
996
446
2,034
870
345
997
874
355
1,037
867
337
Independent variables
2,079
Treatment (yes=1)
Baseline dependent variable values
Baseline past month’s
formal earnings
Baseline past month’s
formal + informal
earnings
Baseline past month’s
total personal income
Baseline past month’s
total public income
Baseline past month’s
private non-earned
income
Baseline past month’s
non-earned personal
income
Baseline past month’s
SSDI + SSI
Chapter 4: Outcomes
4-40
Table 4-23.
Descriptive statistics for dependent measures included in multivariate analyses of
the MHTS impact on earnings and income by treatment and control group
(continued)
All observations
Variables
N
M
SD
Baseline socio-demographics
2,077
45.4
7.7
Age at enrollment
2,079
0.47
0.50
Gender (male=1)
2,079
0.39
0.49
Race (non-white=1)
1,015
1,015
1,015
45.3
0.47
0.39
7.7
0.50
0.49
1,062
1,064
1,064
45.6
0.48
0.38
7.7
0.50
0.49
High school graduate
(yes=1)
Some college (yes=1)
N
Treatment only
M
SD
N
Control only
M
SD
2,078
2,078
0.69
0.18
0.46
0.39
1,014
1,014
0.69
0.19
0.46
0.39
1,064
1,064
0.70
0.18
0.46
0.38
2,075
0.38
0.82
1,015
0.38
0.85
1,060
0.39
0.78
2,076
2,074
0.46
0.46
0.87
0.50
1,015
1,015
0.46
0.46
0.89
0.50
1,061
1,061
0.46
0.46
0.85
0.50
2,077
0.23
0.42
1,015
0.23
0.42
1,062
0.23
0.42
2,079
0.71
0.45
1,015
0.69
0.46
1,064
0.73
0.44
2,079
2,079
110
0.17
72
0.37
1,015
1,015
108
0.17
73
0.37
1,064
1,064
113
0.17
70
0.38
2,079 8,496
Work history and vocational services
3,458
1,015
8,536
3,438
1,064
8,456
3,479
Baseline health status
Fair or poor health (yes=1)
Physically limited in daily
activities (yes=1)
Baseline hospital stays
Baseline emergency room
visits
Diagnosis (affective
disorder=1)
SSA program variables
Months on rolls
Concurrent with SSI
Primary insurance amount
(ln)
Sq. root of reported
earnings in 23 months
pre-baseline
No reported earnings in
23 months prebaseline (yes=1)
Worked in past 2 years
(pre-baseline)
Trial work period end date
0-3 years postrecruitment
Trial work period end date
0-10 years ago
Had active ticket ever
2,078
1,473
4,910
1,015
1,430
4,733
1,063
1,515
5,084
2,079
0.79
0.41
1,015
0.80
0.40
1,064
0.78
0.41
2,074
0.30
0.46
1,012
0.28
0.45
1,062
0.31
0.46
2,078
0.02
0.15
1,015
0.03
0.17
1,063
0.02
0.13
2,078
2,078
0.14
0.14
0.35
0.35
1,015
1,015
0.15
0.14
0.36
0.35
1,063
1,063
0.14
0.14
0.34
0.35
Covariates used in the regression analyses. Regression estimates of the effects of the MHTS
intervention on the dependent variables came from regressions that included a selection of baseline
covariates as additional statistical controls. As is indicated in Appendix 4D, covariates selected to
represent the following categories of explanatory factors include the following: socio-demographic
characteristics, baseline health status (including recent use of health or mental health services),
receipt of benefits (both status and history), recent labor market history and work status, and longer-
Chapter 4: Outcomes
4-41
term work history and vocational services. Covariates pertaining to receipt of benefit history, and
some of the covariates relating to history of work and vocational services, were selected because
these were significant predictors of enrollment probability in our pervious analysis on participants‘
decision to participate in the MHTS (Discussed in Chapter 3).
Functional form and estimation method. The choice of the functional form and estimation
methods reflect several considerations about the distribution of the dependent variable. First, the
significant number of participants who had no earnings in the relevant followup period resulted in a
large concentration of zero values for the dependent variables. Second, the presence of small
numbers of very large dependent variable values suggests that heteroscedasticity (i.e., differing
variance) is a strong possibility. Due to the expected heteroscedasticity pattern, it is also very likely
that there is a positive relationship between the expected value of the dependent variable and the
variance of the unobservable random disturbance. Third, in the presence of heteroscedasticity and
the right-skewed distribution of observed positive values for the dependent variable, a modeling
approach that involves transformation of the dependent variable values (e.g., a two-part model with
a logarithmic transformation, or a one-part model with a square root transformation) raises practical
difficulties in generating estimates of marginal effects via a retransformation process.
Given these considerations, the most straightforward approach for using a regression model for
generating MHTS impact estimates is to use the generalized linear model (GLM) technique with a
logarithmic link function and a gamma error distribution. This model allows the variance of the
unobservable disturbance to increase with the square of the expected value of the dependent
variable. This method has previously been applied in several recent studies that analyze earnings for
persons with serious mental disorders.
One or more regression analyses did not include a small number of participants because of ―don‘t
know‖ or ―refused‖ responses to questions about income amounts from specific income sources. A
few other cases were dropped because they had missing values for baseline covariates. Specifically
for regressions using the 12-month followup periods of earnings, a few other cases were dropped
because the participant did not complete any quarterly interviews during the relevant 12-month time
period.
Since the dependent variables were averaged across all complete interviews for the relevant followup
periods, and since some participants did not complete all interviews, it is important to consider
variations among participants in the number of quarters over which the average earnings values were
computed as a possible source of heteroscedasticity. The investigators expected a correlation
Chapter 4: Outcomes
4-42
between the variance of the errors in our regression estimates and the number of interviews over
which the average dependent variable values were computed.
The number of interviews used to compute the average values for the dependent variables was
generally close to the maximum. For example, for formal earnings, for the two-year study (with a
maximum of 8 such interviews), the mean was 7.09; for the 12-month (first year) interviews (with a
maximum of 4 such interviews), the mean numbers of interviews were 3.59 and 3.65 for the first and
second 12-month periods respectively. Mean numbers of interviews for formal-plus-informal
earnings were the same.
Similar observations pertain to our five dependent variable measures of income over the 24-month
study period. The mean numbers of interviews used for averaging were as follows: (1) SSDI-SSI
6.95; (2) All Public-Sector 6.92; (3) Private Sector Non-earned 7.10; (4) Total Not Earned 7.12;
(5) Total Personal Income 6.83. Differences between these figures and the maximum of 8 again
were relatively small.4
Accordingly, it is reasonable to conclude that variation in numbers of completed interviews was not
an important source of heteroscedasticity. Therefore, there was no need for further adjustments for
heteroscedasticity beyond the gamma distribution assumption for the error term and the use of
robust variance estimates.
Regression results. Regression results are reported for 4 different specifications of covariates:
regressions with only the treatment group (1 = yes, 0 = no) as an explanatory variable (Model 1),
regressions with the treatment group and the baseline value of the dependent variable as the
explanatory variables (Model 2), regressions that add the other categories of covariates (Model 3),
and regressions that replicate Model 3 but exclude all covariates with a p-value for their estimated
average marginal effect that is > 0.2.
In presenting the regression results, the emphasis is on the marginal effects of the treatment variable
(the MHTS dummy) and the selected covariates defined above. These marginal effects, computed
for each participant using the observed covariate values, were then averaged over all participants in
4
As these mean figures suggest, the numbers of beneficiaries with very small numbers of interviews used for computing our dependent variable values
were very small. For 24-month followup data, for formal earnings only 2.54 percent of beneficiaries had fewer than 3 interviews with available data;
for informal earnings the corresponding figure was 2.64 percent. For the 12-month followups, the percentages of beneficiaries with only one
interview with usable earnings data ranged from 2.30 percent to 3.20 percent. For our 5 income variables measured over the 24-month followup,
corresponding percentages of beneficiaries with less than 3 usable interviews were as follows: (1) SSDI-SSI 3.63; (2) All Public-Sector 3.69;
(3) Private Sector Non-earned 2.54; (4) Total Not Earned 2.31; (5) Total Personal Income 4.03. Of course, excluding from our analyses all
beneficiaries who completed no followup interviews beyond the first quarter is consistent with these low percentages.
Chapter 4: Outcomes
4-43
the analysis. Thus, the estimated MHTS impact is the average difference between the predicted
values of the dependent variables (for values of 0 vs. 1 for the treatment variable) with the average
computed over the entire weighted study sample of participants. Variance estimates for all marginal
effects were obtained using robust methods and allow for clustering at the site level.
Table 4-24 reports the estimated MHTS effects on earnings for Models 1 to 3. In all cases, these
estimated effects were positive and significant. Controlling for the baseline dependent variable value
and on the values of the other baseline covariates, results in substantial increases in the estimated
size of the MHTS effects across all three models. There was also a clear tendency for the estimated
effects to be larger for the second year of the study compared to the first year of the study. For
example, the results with Model 3 indicate that the estimated MHTS effect in the second year was
equal to or more than $100 per month in average earnings compared to estimates of $50 and $63 per
month in the first year.
Table 4-24.
Estimates of average marginal MHTS effects on earnings
Dependent variables
Avg. formal earnings
(Q1–Q8)
Avg. formal earnings
(Q1–Q4)
Avg. formal earnings
(Q5–Q8)
Avg. formal + informal
earnings (Q1–Q8)
Avg. formal + informal
earnings (Q1–Q4)
Avg. formal + informal
earnings (Q5–Q8)
N
Model 1
Avg
MEff p-value
Model 2
Avg
MEff p-value
N
2,051
51.17 <0.001
2,051
56.76 <0.001
2,051
72.35
<0.001
2,044
30.87
0.001
2,044
35.32
2,044
50.12
<0.001
1,972
73.16
0.001
1,972
78.00 <0.001
1,972 100.00
<0.001
2,046
52.54 <0.001
2,046
67.31
0.001
2,046
80.70
<0.001
2,039
32.51
0.001
2,039
52.07
0.031
2,039
62.60
<0.001
1,967
74.07
0.001
1,967
81.09
0.001
1,967 102.19
<0.001
0.001
N
Model 3
Avg
MEff
p-value
Model 1: GLM model (gamma family, log link) with treatment dummy as only independent variable.
Model 2: GLM model (gamma family, log link) with treatment dummy and baseline earnings as only independent variables.
Model 3: GLM model (gamma family, log link) with treatment dummy, baseline earnings, and 20 other covariates as independent
variables.
Table 4-25 shows the estimated MHTS effects on the five income measures. Most of these estimates
were not statistically significant; however, the estimates of small negative MHTS impacts on SSDI +
SSI income did have p values < 0.1. Estimates for broader measures of non-earned income,
however, did not support the hypothesis that MHTS treatment reduces receipts of non-earned
income. Positive and significant estimated MHTS effects on total personal income were clearly the
result of the positive MHTS effects on earnings; the fact that these estimates were somewhat smaller
Chapter 4: Outcomes
4-44
than the estimated earnings effects suggests the possibility of negative MHTS impacts on nonearned income but other results in the table give limited support to that hypothesis.
Table 4-25.
Estimates of average marginal MHTS effects on income
Dependent variables
Avg. SSDI + SSI (Q1–Q8)
Avg. public income
(Q1–Q8)
Avg. private non-earned
income (Q1–Q8)
Avg. non-earned personal
income (Q1–Q8)
Avg. formal + informal
earnings (Q5–Q8)
Model 1
Avg
N
MEff p-value
2,010 -5.88 0.645
Model 2
Avg
N
MEff p-value
2,010 -11.99 0.071
Model 3
Avg
N
MEff
p-value
2,010 -11.19 0.089
2,004
4.25
0.807
2,004 -17.98
0.096
2,004
-14.51
0.129
2,045
-0.30
0.984
2,045
80.85
0.892
2,045 865.96
0.555
1,992
3.14
0.867
1,992 -14.23
0.406
1,992
-12.55
0.378
1,985
61.13
0.004
1,985
0.006
1,985
52.14
0.002
61.60
Model 1: GLM model (gamma family, log link) with treatment dummy as only independent variable.
Model 2: GLM model (gamma family, log link) with treatment dummy and baseline earnings as only independent variables.
Model 3: GLM model (gamma family, log link) with treatment dummy, baseline earnings, and 20 other covariates as independent
variables.
Estimated average marginal effects of all other covariates are reported in Appendix Table 4D (for
formal earnings), Appendix Table 4E (for formal + informal earnings,) and in Appendix Table 4F
(for the five income dependent variables). These tables also report analogous results for the MHTS
treatment dummy that were already shown (for Model 3) in Tables 2 and 3, as well as the MHTS
results when insignificant covariates were deleted from Model 3. It is clear that the estimated MHTS
effects were quite stable to the exclusion of covariates that were least significant in the full Model 3.
Health and Quality of Life
While the primary outcomes of interest were vocational, focusing on employment and earnings, the
rich nature of the treatment was also expected to affect health and functioning. Access to SMM and
other behavioral health services was a significant part of the treatment package. For those
participants who took advantage of the services, improvements were expected.
Health status. This study reports on two health status scores—the Mental Component Score
(MCS), which provides a summary of the participant‘s mental health in the past 4 weeks, and the
Physical Component Score (PCS), which provides a summary of the participant‘s physical health in
the past 4 weeks. Population norms associated with the MCS and PCS have a mean of 50 with a
Chapter 4: Outcomes
4-45
standard deviation of 10. Thus, higher scores indicate better mental or physical health. As scores
deviate from the population mean of 50, they indicate better (above 50) or worse (below 50) health
than persons in the normal population. The norms were based on the 1998 U.S. population (Ware,
Kosinski, Turner-Bowker, & Gandek, 2002).
Table 4-26 presents a summary of the MCS and PCS at baseline and study exit for both the
treatment and control groups. As can be seen in the table, mental health scores are well below the
normal population between 1 and 1 ½ standard deviations, indicating that the mental health of
participants in the study was quite poor. However, the results indicate that average MCS improved
in the treatment group (from 35.83 to 38.85) during their 24 months in the study, while the average
MCS of participants in the control group changed very little (from 35.96 to 35.92). The difference in
change scores (3.02 vs. -0.04) was statistically significant (p-value < 0.001).
Table 4-26.
Mental and physical health status at baseline and study exit by treatment and
control group
Variable
Mental health
Mean
SD
Physical health
Mean
SD
Treatment
(N=1,004)
Baseline
Exit
d
Control
(N=1,051)
Baseline
Exit
d
p-value
35.83
13.08
38.85
13.37
3.02
35.96
13.00
35.92
13.27
-0.04
<0.001
44.27
11.90
43.13
11.69
-1.14
43.96
11.86
42.92
12.21
-1.04
0.924
Further analyses of the average MCS scores (see Appendix 4G) of participants reveal that the
treatment versus control group differences were also statistically significant (at least at the.05 level)
for the following key subgroups:

Participants in the age groups 18 to 34 and those 35 and older,

Males and females,

High school graduates and participants with more than a high school education, and

Participants diagnosed with an affective disorder as well as participants diagnosed with
schizophrenia.
The only subgroup that did not show any significance between treatment and control at a
significance level of.05 from baseline to study exit were participants with less than a high school
Chapter 4: Outcomes
4-46
education (p-value = 0.060; refer to Appendix 4G for statistics); however, increasing the significance
level to.10 reveals marginally significant differences for this subgroup.
Average physical health scores were comparatively higher than the mental health scores. While still
falling below normal population mean of 50, physical health scores for both the treatment and
control groups remained within a single standard deviation of the normal population. The average
scores dropped slightly between both the treatment and control groups between the Baseline and
final Followup interviews. Thus, the physical health of participants in the study declined slightly in
the 24 months of study participation. The difference in change scores between the treatment and
control groups (or among any subgroup—see Appendix 4G) was not significant (p-value = 0.924).
Predicting mental health status. A stepwise linear regression model provided a means to obtain
estimates for the mental health score at study exit. MHTS treatment showed a highly significant
positive impact on mental health status (see Table 4-27). Participants with higher baseline mental
health scores, those with an affective disorder, and participants with an active ticket also reported
better mental health at study exit. While age had a negative effect, participants living in areas with
higher population density also reported better mental health at exit.
Table 4-27.
Linear regression for mental health status at study exit (for all participants)
Variable
Treatment dummy
Age
Ever had active ticket
Diagnosis
Baseline mental health (SF-12)
Population density
Number of ER visits at baseline
Estimate
2.87
-0.44
1.86
3.06
0.46
0.62
-0.58
p-value
<0.001
0.209
0.011
<0.001
<0.001
<0.001
0.089
R-squared = 0.2665
Table 4-28 presents stepwise linear regression results for the mental health status at study exit for
treatment group participants only. As expected, baseline mental health score was a highly significant
predictor of mental health status at study exit. Receiving general medical care and engagement in
SMM also had significant and positive effects on mental health at exit. Participants with an affective
disorder and those who lived in areas with higher population density reported better mental health at
exit.
Chapter 4: Outcomes
4-47
Table 4-28.
Linear regression for mental health status at study exit (for treatment group
participants only)
Variable
Received general medical care
SMM engagement category
Diagnosis
Baseline mental health (SF-12)
Population density
Estimate
2.50
1.51
2.25
0.49
0.57
p-value
0.047
0.020
0.008
<0.001
<0.001
R-squared = 0.2910
Quality of life. The Modified Lehman Quality of Life Inventory provided the single item that
measured quality of life in the study. As noted in Chapter 2, individuals who obtain competitive
employment over time report higher quality of life. In this study, interviewers asked participants the
single item on general life satisfaction during the Baseline interview and then again during the final
Followup interview. Table 4-29 presents the study results. General life satisfaction reported by
participants in the treatment group improved significantly over that reported by participants in the
control group (p-value < 0.001).
Table 4-29.
Satisfaction with life at baseline and study exit by treatment and control group
Variable
Life satisfaction
Mean
SD
Treatment
(N=1,004)
Baseline Followup
3.78
1.55
4.22
1.51
Control
(N=1,051)
Baseline Followup
d
0.44
3.82
1.53
4.01
1.58
d
p-value
0.19
0.0002
Appendix 4G also presents tabled values of general life satisfaction by relevant subgroups. In
summary, the results indicate that significant differences always favored the treatment group. The
following subgroups show significant differences (p-value <.05) favoring the treatment group:

Participants 35 years of age and older,

Males and females,

Participants diagnosed with an affective disorder, and

Participants with a high school education and those with more than a high school
education.
No significant findings between the treatment and control groups were noted in any of the
remaining subgroups, including participants under the age of 35 or participants with schizophrenia.
Chapter 4: Outcomes
4-48
Predicting quality of life. This section presents an analysis of the MHTS effects on the quality of
life scores at study exit. An ordered logistic regression technique was used to estimate the categorical
outcome quality of life, where some outcomes are better (or valued higher) than others. Participants
reported on their level of satisfaction with their life in general using Lehman’s terrible—delighted
scale. The scale ranges from a value of 1 (terrible) to 7 (delighted). In order to provide an easier
interpretation of the findings, the three lowest scores (terrible, unhappy, mostly unsatisfied) and the
three highest scores (mostly satisfied, pleased, delighted) were aggregated. Results were similar to the
regression estimates using the seven-point scale.
Table 4-30 presents the results of ordered logistic regression estimates. The results suggest that the
MHTS treatment had a significant positive impact on quality of life at study exit. MHTS participants
were 6.4 percent less likely to report feeling “terrible, unhappy, or mostly unsatisfied” about their life
in general, and they were 7.5 percent more likely to report feeling “mostly satisfied, pleased, or
delighted.” In addition, baseline mental and physical health status, diagnosis of an affective disorder,
and having an active ticket were all statistically significant with positive effects on the quality of life
outcome measure. Local unemployment rate and age were significant as well, and had a negative
relationship with the quality of life at study exit. As with previous models, being in the treatment
group was associated with higher quality of life.
Table 4-31 presents results for the ordered logit estimates for the quality of life score at study exit
for treatment group participants only. Having better mental and physical health at baseline were
both significant with positive impacts on the quality of life score. Age was significant with negative
effects on the quality of life score. Engagement with SMM was highly significant with a positive
effect on quality of life. Clients receiving social skills training or receiving case management were
more likely to report a lower quality of life score.
Chapter 4: Outcomes
4-49
Table 4-30.
Ordered logistic regression estimates for quality of life (all participants)
Dependent variable
Treatment dummy
Baseline mental health (SF-12)
Baseline physical health (SF-12)
Gender
Worked in last 2 years (prebaseline)
Diagnosis
Age
Less than HS graduate vs. more
than HS graduate
HS graduate vs. more than HS
graduate
Unemployment rate
Months on rolls
Number of ER visits at baseline
Number of hospital stays at
baseline
Population density
Not receiving SSI (SSDI only)
Ever had active ticket
Table 4-31.
Terrible, unhappy,
and mostly
unsatisfied
Estimate p-value
-0.064 <0.001
-0.013 <0.001
-0.004 <0.001
0.027
0.145
-0.005
0.801
Mixed
Estimate p-value
-0.011
0.003
-0.002 <0.001
-0.001
0.001
0.004
0.153
-0.001
0.805
Mostly satisfied,
pleased, and
delighted
Estimate p-value
0.075 <0.001
0.015 <0.001
0.004 <0.001
-0.031
0.144
0.006
0.801
-0.053
0.003
0.043
0.015
0.030
0.161
-0.008
0.001
0.004
0.029
0.018
0.021
0.061
-0.004
-0.047
0.015
0.007
0.139
-0.020
0.319
-0.004
0.374
0.024
0.327
0.020
0.001
-0.003
-0.006
0.001
0.597
0.803
0.644
0.003
0.001
0.001
-0.001
0.004
0.599
0.803
0.646
-0.023
0.001
0.003
0.007
0.001
0.597
0.803
0.644
0.005
-0.022
-0.051
0.331
0.364
0.047
0.001
-0.004
-0.008
0.340
0.437
0.065
-0.006
0.026
0.060
0.331
0.376
0.047
Ordered logistic regression estimates for quality of life (treatment group
participants only)
Dependent variable
Treatment dummy
Baseline mental health (SF-12)
Baseline physical health (SF-12)
Gender
Worked in last 2 years (prebaseline)
Diagnosis
Age
Less than HS graduate vs. more
than HS graduate
HS graduate vs. more than HS
graduate
Unemployment rate
Chapter 4: Outcomes
Terrible, unhappy,
and mostly
unsatisfied
Estimate p-value
Mixed
Estimate p-value
Mostly satisfied,
pleased, and
delighted
Estimate p-value
-0.013
-0.004
0.030
-0.002
0.001
<0.001
0.218
0.941
-0.004
-0.001
0.010
-0.001
0.001
0.001
0.217
0.941
0.017
0.006
-0.040
0.003
<0.001
<0.001
0.215
0.941
-0.023
0.003
0.036
0.420
0.030
0.373
-0.008
0.001
0.009
0.425
0.042
0.242
0.031
-0.005
-0.046
0.420
0.029
0.347
-0.015
0.597
-0.005
0.618
0.020
0.602
0.014
0.067
0.005
0.081
-0.019
0.067
4-50
Table 4-31.
Ordered logistic regression estimates for quality of life (treatment group
participants only) (continued)
Dependent variable
Months on rolls
Number of ER visits at baseline
Number of hospital stays at
baseline
Population density
Not receiving SSI (SSDI only)
Ever had active ticket
SE engagement score
SMM engagement category
Received medication services
Received general medical care
Received substance abuse
treatment
Received housing services
Received family counseling or legal
assistance
Received social skills training
Received financial assistance
training
Received case management
Average site fidelity score
Number of clients served at site
Treatment dummy
Baseline mental health (SF-12)
Terrible, unhappy,
and mostly
unsatisfied
Estimate p-value
0.001
0.719
-0.016
0.297
0.008
0.638
Mixed
Estimate p-value
0.001
0.719
-0.005
0.304
0.003
0.639
Mostly satisfied,
pleased, and
delighted
Estimate p-value
0.001
0.719
0.021
0.297
-0.010
0.638
-0.004
0.057
-0.022
0.003
-0.071
-0.054
0.028
-0.004
0.557
0.116
0.499
0.177
0.001
0.190
0.477
0.906
-0.001
0.013
-0.008
0.001
-0.024
-0.017
0.009
-0.002
0.559
0.027
0.548
0.192
0.003
0.185
0.476
0.908
0.006
-0.070
0.030
-0.004
0.095
0.072
-0.038
0.006
0.557
0.090
0.512
0.178
0.001
0.186
0.476
0.906
-0.043
-0.040
0.235
0.274
-0.018
-0.016
0.330
0.366
0.061
0.056
0.263
0.301
0.115
-0.019
0.008
0.604
0.018
-0.007
0.002
0.639
-0.133
0.025
0.002
0.614
0.074
-0.001
-0.002
0.052
0.795
0.428
0.020
0.001
-0.001
0.028
0.795
0.433
-0.094
0.002
0.003
0.042
0.795
0.428
-0.013
0.001
-0.004
0.001
0.017
<0.001
Alcohol or drug use. Table 4-32 shows the change in the percentage of study participants meeting
the Addiction Severity Index (ASI) cutoff score for alcohol use at baseline and exit. Among the
treatment group participants, while 9.1 percent reported an alcohol use problem at baseline, this
figure dropped to 6.2 percent at exit. Similarly, while 8.3 percent of the control group participants
reported an alcohol use problem at baseline, this figure dropped to 5.2 percent at exit. Both groups
had almost a one-third reduction in alcohol use at the time they exited the study. However, the
difference between treatment and control groups was not statistically significant (p-value = 0.967).
Table 4-32 also shows the change in the percentage of study participants meeting the ASI cutoff
score for drug use at baseline and exit. Although the percent of participants reporting drug problems
Chapter 4: Outcomes
4-51
increased among the treatment group, there is statistically no significant difference between the two
groups in terms of change in drug use problems (p-value = 0.238).
Table 4-32.
Alcohol and drug use at baseline and study exit by treatment and control group
Variable
Alcohol use score > 0.17
n
%
Drug use score > 0.16
n
%
Treatment
(N=1,004)
Baseline
Exit
91
9.1
62
6.2
39
3.9
53
5.3
Control
(N=1,051)
Baseline
Exit
d
d
p-value
-2.9
86
8.3
54
5.2
-3.1
0.967
1.4
47
4.5
47
4.4
-0.1
0.238
NOTE: Weighted percents may not be consistent with unweighted counts.
Table 4-33 shows that treatment and control group participants reporting any drug use during the
last 30 days went down from 52.0 percent and 51.1 percent to 37.8 percent and 38.4 percent,
respectively. While the change from baseline to exit is significant for both groups, there is no
statistically significant difference between treatment and control (p-value = 0.5340).
Table 4-33.
Any drug use during the past 30 days at baseline and study exit by treatment and
control group
Variable
Any drug use past 30 days
n
%
Treatment
(N=1,004)
Baseline
Exit
522
51.95
379
37.75
d
14.20
Control
(N=1,051)
Baseline
Exit
536
51.09
405
38.39
d
12.70
p-value
0.5340
NOTE: Weighted percents may not be consistent with unweighted counts.
The MHTS was designed to provide employment supports, SMM, and insurance supports to
improve participants‘ employment and mental health outcomes. Most MHTS study sites were not
equipped to provide substance use treatment. Thus, consistent with our earlier expectations, change
in alcohol or drug use does not indicate statistically significant differences between the treatment
and control groups.
Chapter 4: Outcomes
4-52
Discussion
The MHTS was designed to test the fundamental value of providing evidence-based mental health
treatment and employment supports to SSDI beneficiaries to improve their employment, health, and
functioning. However, the primary emphasis was on employment. The basic components of the
treatment included IPS SE services, SMM, and other behavioral health services—all of which were
paid for with a comprehensive health insurance package.
Employment
The data presented in this chapter clearly specify that the employment rates of participants who
received the treatment intervention were significantly and meaningfully improved over participants
in the control group. When viewed in the context of all participants in the intent-to-treat analysis,
the treatment group participants were more likely to find employment, worked more months,
worked more months at study exit, and worked more hours per week than did control group
participants. These findings were consistent across different demographic, clinical, Social Security
program title (SSDI and SSI), and, with few exceptions, were consistent across sites as well. These
findings only partially extended to the subgroups of participants who worked any job or at least one
competitive job.
Overall employment and competitive employment were generally consistent with previous research
on IPS SE showing that SSDI beneficiaries achieve much better than in other vocational
interventions or in no intervention at all via a treatment-as-usual design. The fact that participants in
the treatment group were employed for more months than were participants in the control group
accords with previous IPS research (Bond, Drake, & Becker, 2008). The parsimonious interpretation
is that IPS helps people to find jobs and to keep them, and this is a reasonable interpretation of the
findings from the MHTS.
The findings were, however, surprising in one aspect, namely that the control group participants did
remarkably well. Despite the fact that participants in the treatment group did significantly better than
those in the control group, the overall paid and competitive employment rates for the control group
participants were much higher than those of SSDI beneficiaries in previous IPS studies (Bond, Xie,
& Drake, 2007). The SSDI beneficiaries in the MHTS were different from those in previous IPS
studies in that most were not attending community mental health centers prior to the study and self-
Chapter 4: Outcomes
4-53
selected for participation based on letters from SSA and introduction meetings that clarified that the
study was for beneficiaries who wanted to return to work. Therefore, it is reasonable to assume that
they were highly motivated to work.
Comparison of MHTS employment outcomes with the IPS literature. While the overall
findings unambiguously show a strong statistical advantage to the treatment condition over the
control condition on employment outcome measures, one might also ask more specifically how
strong these findings are, and how they compare to the general literature on the effectiveness of IPS.
One comparison would be to the aggregated findings from 11 randomized controlled trials (RCTs)
of IPS reviewed by Bond et al. (2008).
Several important caveats are in order for comparisons between the MHTS and this earlier IPS
literature: the samples differ in important ways, as do the interventions themselves. Regarding the
samples, most of the RCTs of IPS reviewed by Bond et al. (2008) enrolled participants who were
already clients in the public mental health system at the time of enrollment and only a minority
(approximately 40%) were SSDI beneficiaries (Bond, Xie, & Drake, 2007). Regarding the
interventions, the participants in the 11 RCTs generally received mental health case management and
a range of other behavioral services that were unevenly offered to MHTS treatment study
participants. On the other hand, participants in these earlier studies did not receive the full package
of benefits the MHTS treatment participants did, including financial assistance for out-of-pocket
medical expenses, SMM, assistance from NCCs, and suspended CDRs.
Regarding differences in outcomes, one difference between the current study and the IPS literature
is the proportion of treatment participants in the current study who obtained noncompetitive paid
jobs (7.9%) as their highest level of paid employment. In most of the prior IPS studies the rate is
probably lower. For example, in one prior IPS study, 4 percent of treatment participants obtained
casual labor and 1 percent were placed in a sheltered workshop as their highest level of paid
employment (Bond, Salyers, et al., 2007).
In terms of overall employment rates for the treatment group in the current study—61 percent in
paid employment and 53 percent in competitive employment—these rates fall with the midrange for
the Bond et al. (2007) review, which reported a mean competitive employment rate of 61 percent
across 11 studies. In the current study, the average time to the first job for treatment participants
who obtained work was 7.7 months, compared to 4.6 months for seven studies in the Bond et al.
(2007) review, suggesting a much longer delay to the first job in the current study. Unlike the
findings in the Bond et al. (2007) review article, which found IPS participants achieving their first
Chapter 4: Outcomes
4-54
job significantly sooner than controls, time to first job did not differ between treatment and controls
in the current study. It is reasonable to speculate that the added demands of testing required of
treatment participants (e.g., General Medical Exam, cognitive testing) may have contributed to this
delay, although it is also possible that the engagement process was more extended in the current
study, given that most beneficiaries were new to the study site.
In the current study, the total sample of treatment participants averaged 6.2 months of employment
over the 2-year period, which exceeds the average of 5.6 months (24.2 weeks) of competitive
employment for seven studies in the Bond et al. (2007) review. Similar comparisons were made for
participants who obtained work. In the current study, treatment participants who obtained work
averaged 10.3 months of employment over the 2-year period, which also exceeds the average of 8.9
months (38.4 weeks) of competitive employment for seven studies in the Bond et al. (2007) review.
Finally, treatment participants had greater job satisfaction than controls for their main job (assessed
at followup), a difference found in some, but not all, of the previous IPS studies examining job
satisfaction.
Another key difference favoring the current study over prior IPS studies was the significantly better
non-vocational outcomes for treatment participants, as discussed below. Controlled studies
consistently fail to show significant improvement in non-vocational domains for IPS clients,
compared to clients receiving traditional vocational services (Burns et al., 2009; Drake et al., 1999;
Drake, McHugo, Becker, Anthony, & Clark, 1996; Gold et al., 2006; Latimer et al., 2006; Lehman et
al., 2002; Mueser et al., 2004; Twamley, Narvaez, Becker, Bartels, & Jeste, 2008). One major
difference between the current study and these earlier IPS studies is that none of these earlier studies
included SMM or other behavioral health services that were provided solely to the participants in the
treatment intervention. Since the difference between the MHTS and other IPS studies was the
provision of the non-IPS supports in the MHTS, it seems reasonable that these other supports were
primarily responsible for the improvement in non-vocational outcomes.
To summarize the comparisons between the outcomes in the current study with the IPS literature,
the treatment condition generally attained employment outcomes comparable to those reported in
the literature for IPS participants. Time to first job was the one main exception.
The comparisons to the IPS literature raise the question about the consequences if the unique
features of the MHTS treatment model (e.g., financial assistance for out-of-pocket medical expenses,
SMM, assistance from NCCs, and suspended CDRs) had not been provided. The short answer is
Chapter 4: Outcomes
4-55
that the design of the MHTS does not permit a statistical answer to the dismantling question,
because these features were offered as a package and therefore were completely confounded with
IPS. However, the literature suggests that when clients do not receive vocational assistance (i.e.,
IPS), they do not increase employment outcomes, regardless of the excellence of mental health
treatment, so this implies that vocational assistance is a necessary critical ingredient (Bond, 1998).
However, it is also true that vocational assistance is much less effective when it is not well integrated
with mental health treatment (Cook et al., 2005).
Site differences in employment. Employment findings were consistent across most of the sites in
the MHTS. Altogether, 21 of 23 showed better employment outcomes for the treatment condition
than control. Of the two sites that did not show a clear advantage for the treatment group, one had
an unusually high rate of competitive employment in the control group (52%), probably due to
extreme treatment drift, and the other had an unusually low rate of competitive employment in the
treatment group (29%), almost certainly due to the site leader‘s unwillingness to follow the model,
due to disagreement with the model. MHTS investigators considered excluding the latter site from
the study at several points, but retained it as an example of extreme model changes. Program
planners often believe that they can change the IPS model to do it better (Menear et al., 2011). This
case example illustrates the most common outcome when a program leader makes fundamental
changes in the IPS model; the outcomes suffer.
Predicting employment outcomes from participant characteristics. The predictors of
employment among treatment group participants are entirely congruent with the literature on SE.
Younger clients with serious mental illness are considered to have better vocational potential,
particularly those in the early phases of illness who have not yet been socialized into disability
(Rinaldi et al., 2010). Poor physical health status as a predictor has not received extensive attention
in studies. The SF-12 measure assesses perceived impairment due to physical illnesses rather than
objective measures of physical illness, which was examined in the ―reasons for not working‖ section.
This finding regarding inability to obtain employment adds to growing concerns about the physical
health status of people with serious mental illness. Recent earnings and recent work history may be
proxies for capacity, readiness, motivation, or a complex mixture of these and other constructs. As
expected, recent work history was a strong predictor of obtaining employment and of amount of
work. The amount of work among participants who obtained at least one job is related to work
history, which is a predictor in many studies (Bond & Drake, 2008; Campbell et al., 2010). Work
history undoubtedly captures explanatory value by encompassing capacity, motivation, and other
constructs.
Chapter 4: Outcomes
4-56
More important for policy purposes, engaging in IPS SE services predicted obtaining employment.
The study findings clearly support the view that SSDI beneficiaries benefit from high-quality
employment services in pursuit of the goal of employment.
Engagement in SMM was a significant predictor of the number of months of employment in the
treatment group in the regression analysis. This finding can be interpreted to mean that participants
who were judged as less engaged with their prescriber may have had less contact with the NCC—
and by extension, with other treatment services at the study site—and achieved fewer months in
employment. It is unclear whether better medication management per se accounted for this
association, because otherwise one would have expected a stronger association between engagement
in SMM and change in symptoms. The correlation between engagement in SMM and improvement
in self-reported psychiatric symptoms was weak (r =.08).
An alternative explanation is that engagement in SMM is a proxy for receipt of integrated services at
the study site. If this speculation is correct, these findings would be consistent with research
suggesting the critical role of integrated services on employment outcomes (Cook et al., 2005; Drake,
Becker, Bond, & Mueser, 2003). Additional analyses of the existing data will likely reveal a clearer
understanding of these relationships.
Types of occupations. The large preponderance jobs held by study participants (just under 75% of
all jobs held at any time) were in two sectors: Service and Sales/Office. There are only modest
proportional differences in the occupations held by treatment group participants compared to
controls. Although these differences were not tested statistically, the clinical importance of any
differences appears negligible; therefore, it is reasonable to conclude that the kinds of jobs obtained
by treatment group participants were similar in proportion to those obtained by participants in the
control group. The data also reveal that both diagnosis and education affect the types of jobs
participants obtain, but there are no obvious differences in attainment of occupational type between
the treatment and control groups within diagnosis or within educational level.
In terms of the types of jobs obtained by MHTS study participants, the distribution of occupational
types mirrors that commonly obtained in studies of SE clients with severe mental illness, as reported
in the literature spanning the last two decades, including recent studies (Becker, Bebout, & Drake,
1998; Becker, Drake, Farabaugh, & Bond, 1996; Bond, Drake, Mueser, & Becker, 1997; Bond &
Kukla, 2011; Fabian, 1989; Gervey & Kowal, 1994; Heslin et al., 2011; MacDonald-Wilson, Revell,
Nguyen, & Peterson, 1991; Mowbray, McCrohan, & Bybee, 1995; Mueser, Becker, & Wolfe, 2001;
Resnick & Bond, 2001; Shafer & Huang, 1995). Many of the jobs obtained by MHTS participants
Chapter 4: Outcomes
4-57
were entry level and unskilled, reflecting the work experience and skill levels of many participants in
the study sample. While the distribution of jobs might suggest that participants obtained unappealing
jobs, MHTS investigators interpret the findings differently. From the perspective of the IPS model,
the critical question is not occupational type, but job match, as repeatedly documented in the literature
(Becker et al., 1998; Becker et al., 1996; Carpenter & Perkins, 1997; Gervey & Kowal, 1994; Huff,
Rapp, & Campbell, 2008). SE clients who obtain jobs that suit their preferences (which includes type
of occupation, work hours, supervision, location, and many other factors) have greater job
satisfaction and job tenure than those who are poorly matched. Another factor in interpreting
occupational types is the unavoidable reductionism of any classification system. Within a particular
occupational category are both high quality and low quality jobs.
Barriers to employment for participants who achieved little or no success. Near the end of the
data collection period, the NCC, employment specialist, and other team members at each site
identified reasons that they believed were barriers to employment for treatment participants who
had little or no success in obtaining work. This analysis helps us to understand the reasons that some
participants did not work during MHTS. The most prominent barriers clearly were uncontrolled
mental illness, uncontrolled physical health problems, and failure to engage in IPS SE.
The factors with extremely low ratings were also of interest because many were discrepant from the
literature on barriers to employment. Less than 10 percent of the participants were judged to have
barriers related to cognitive problems, lack of services, transportation difficulties, lack of work skills,
lack of social skills, criminal justice system problems, and housing problems. The literature clearly
notes each of these problems as a barrier to work. The lack of prominence in MHTS could be due
to the special characteristics of SSDI beneficiaries (e.g., substantial work histories and other factors
discussed earlier), the services they received through MHTS, the teams‘ failure to recognize some
problems (e.g., cognitive difficulties may be overlooked), or other factors. It should also be
acknowledged that these ratings were based on the perspective of the treatment team, which prior
research indicates differs from client perspectives (Crane-Ross, Roth, & Lauber, 2000), as well as
other data sources.
In addition to these three major barriers, family problems, poorly controlled substance use disorders,
disengagement from SE services, and behavior problems affected greater than 10 percent of the
group. One interpretation of these barriers is as follows: families often rely on the beneficiaries‘
income for support or on the beneficiaries‘ willingness to work in casual labor for the family. Both
issues can lead families to resist the beneficiaries‘ interests in employment because they threaten the
status quo. Some families also recall previous experiences when their relative had a negative work
Chapter 4: Outcomes
4-58
experience or lost disability benefits due to working temporarily. Co-occurring substance use
disorder is very common among people with serious mental illness and may develop or become
exacerbated due to joblessness. Some clients clearly lose jobs if they continue to use substances.
Clients withdraw from services for a variety of reasons. They may decide not to work when they
understand the long-term threat to benefits, when they try to find a job or actually work and find it
more stressful than they imagined, or when they have a relapse of any of their illnesses. Behavior
problems often indicate co-occurring personality disorders that interfere with working.
Earnings and Income
Treatment group participants earned more on average than did control group participants (past
month‘s earnings 1.5 to 1, past three months‘ earnings at study exit 2 to 1). However, these
differences are attributed mostly to the higher employment rate among MHTS treatment group
participants. This differential effect on earnings of participation in treatment was expected as part of
the general hypothesis that the treatment group would achieve better employment outcomes than
controls. The findings are consistent with the overall model of the IPS intervention, which
addressed increased employment and improving earnings.
The study also examined how often study participants had monthly earnings exceeding SGA. As an
exploratory hypothesis, analyses were conducted to examine whether treatment group participants
would have a higher frequency of months above SGA than control group participants. However,
there were very few quarters in which either treatment group participants or control participants
exceeded SGA in the last month of the quarter. In fact, only 6 percent of treatment group
participants and only 3 percent of control group participants exceeded SGA in more than one
quarter (of 8 quarters examined). Thus, it is reasonable to conclude that while the employment rate
for treatment group participants was significantly higher than that for control group participants, the
extent of employment as measured against SGA was not significantly different between the two
groups.
A further exploratory hypothesis was that participants in both groups would ―park‖ just below SGA,
that is, work just under the limit, attaining monthly earnings between 75 percent and 99 percent of
SGA. The results of this analysis were very similar to the preceding analysis: very few of either group
earned between 75 percent and 99 percent of SGA in any month during the study. Only 6 percent of
treatment group participants and only 2 percent of control group participants earned between 75
Chapter 4: Outcomes
4-59
percent and 99 percent of SGA in more than one quarter. Parking behavior occurred at a very low
rate, if at all, in this study.
In summary, economic disincentives (that is, earnings limits) do not appear to have been a potent
factor in how many hours per week study participants worked. MHTS investigators infer that
beneficiary decisions about intensity of employment (i.e., how many hours to work per week) were
dictated by other factors, which might include stamina, lifestyle, and presence of co-occurring
physical conditions.
Beneficiary attitudes towards SSA benefits were also measured using a series of items included in the
Baseline and final Followup interview. The findings suggested the treatment group had a modest
reduction in fears about losing benefits as a result of employment. The control group did not show
much attitude change. The attitude change in the treatment group was probably attributable to the
higher rates of employment and to benefits counseling received as an element of the intervention
offered to the treatment group.
At the end of the MHTS, treatment group members on average had nearly twice the earnings of
members of the control group and more than twice as many were likely to have had earnings in the
prior three months. Monthly income was also higher for the treatment group and for their families.
Income increased faster for the treatment group, reflecting the philosophy and emphasis of the IPS
model of SE on rapid job placement—place and train rather than train and place. While income and
earnings differed significantly between the two groups, mean SSDI benefits remained the same for
both groups, as did benefits from most other social welfare programs. The only treatment group
participants to have an observed significant difference in Social Security benefits were in the
subgroup of participants on SSI. Males and participants with schizophrenia were observed to have
lower income from SSI.
Health and Functioning
The assessments of health and functioning outcomes reveal two prominent findings from the study:
participants in the treatment group exited the study with statistically significant improved mental
health and general satisfaction with life compared to the control group participants. While the
commensurate effect sizes suggest these improvements were modest at best, the patterns of
improvement and the significance extended throughout the entire study sample, with mean
differences always benefitting the treatment group and nearly all subgroups showing a significant
Chapter 4: Outcomes
4-60
difference from the control group. Further analyses of the existing data are needed to assess whether
or not these improvements can be better understood.
The self-reported health status of participants in the study (both treatment and control) is consistent
with the literature. While little normative information about the physical or mental health status
(using the SF-12) of individuals with schizophrenia exists, there are norms for individuals with
depression. Those norms, reported by Ware, Kosinski, Turner-Bowker, and Gandek (2002), suggest
that both the physical and mental health of participants in the study with an affective disorder
approximate the health status of the population with depression. This fact further substantiates the
study findings associated with improved mental health.
Earlier studies found that SE has minimal or no systematic impact on non-vocational outcomes
(Drake, McHugo, et al., 1996; Becker, Bond, et al., 2001; Bond, 2004). However, the intervention
design of the MHTS was substantially different from all of the previous studies examining the
relationship between non-vocational outcomes and SE. The MHTS intervention combined SE
services with SMM and coverage for other behavioral health services. This study finds that the
mental health and general satisfaction with life of the treatment group improved significantly over
the course of the study and over the comparison group.
Alcohol and other substance abuse measures did not change differentially for treatment and control
group participants. It is important to note that substance abuse-related services were not a primary
focus of the intervention. Change in self-reported general health status was also not significantly
different between the two groups, which is also to be expected, since general medical services were
not a differential element for the treatment and control groups. For the most part, both groups had
Medicare coverage during the period of the MHTS and had comparable access to general medical
services. Treatment participants did have access to counseling regarding which Medicare benefits to
obtain and received reimbursement for Parts B and D, but general medical services were not an
explicit element of the treatment group services.
Chapter 4: Outcomes
4-61
Chapter 5
Implementation of Supported Employment and Other Behavioral
Health and Related Services
The primary treatment intervention components of the Mental Health Treatment Study
(MHTS) offered access to systematic medication management (SMM), Individual Placement and
Support (IPS) supported employment (SE), other behavioral health (OBH) and related services, and
Nurse Care Coordinator (NCC) services. This chapter describes implementation of the IPS SE, and
OBH and related services component of the intervention. The description includes an assessment of
several different aspects of implementation. Measurement at the site level assessed the extent to
which the study sites implemented SE services as intended. That is, did the study sites develop and
provide the kind of SE services that are consistent with the IPS model of demonstrated
effectiveness? The relevant term for this level is “program-level fidelity.” Measurement at the
individual level assessed the extent to which treatment group participants received the array of SE
assistance and OBH and related services expected in a program that is faithful to the IPS model. The
relevant term for this level is “beneficiary-level fidelity.” A third aspect of implementation assessed
in the study, also at the beneficiary level, was the extent of active beneficiary involvement in the
intervention offered by the study site. The relevant term for this is “engagement.”
The primary focus of this chapter is on the fidelity of the services implemented in the MHTS to the
planned IPS service model. Even though most of the sites were chosen because they were already
implementing or seeking to implement IPS, a wealth of field experience suggests that, in routine
practice, mental health programs often drift from close adherence to a model, even when program
leaders seek to implement the model as intended (Becker et al., 2006). In addition, agency leaders
commonly assume they are implementing an evidence-based practice when their actual practice
departs from the defined model (Bond et al., 2011).
Thus, a primary objective of this chapter is to convey how closely the 23 study sites, individually and
as a group, adhered to the principles of the IPS model of SE and offered treatment group
participants the wide range of OBH and related services needed to improve their employment,
health, and quality of life. Secondary objectives include an understanding of the rates of receipt of
services, consistency in delivery of the services, where services occurred, and the relationship
between fidelity in the MHTS and employment rates.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-1
Background
Fidelity refers to adherence to the principles of an evidence-based program model. A fidelity scale is
an instrument used to assess adherence to the specific model (Bond, Evans, Salyers, Williams, &
Kim, 2000). In the MHTS, IPS fidelity referred to the degree of adherence to the evidence-based
IPS model of SE within and across the multisite study. IPS was a primary focus of the intervention,
thereby bringing fidelity to the forefront as an important consideration in the overall assessment of
the study implementation. It is possible to have study sites that meet the standards of high programlevel fidelity, but fail to provide all of the required SE or behavioral health and related services. It is
also possible to have high program-level fidelity without all treatment group participants fully
engaged in the service components of the intervention. High program-level fidelity sites may not
obtain optimal outcomes if they do not engage all or most treatment group participants fully or if
many participants do not receive the full array of services. For these reasons, it is important to assess
program-level fidelity, individual-level fidelity, and engagement to determine the extent of
implementation of SE and OBH and related services.
There are four main reasons for assessing fidelity in a multisite randomized controlled trial. The first
is to facilitate communication to readers of reports about the nature and degree of implementation of an
intervention under investigation. The importance of providing a basic description of adherence to
model implementation is evident from a review of the history of model dissemination. Within the
community mental health field, the need for systematic methods to assess fidelity became clear in
the 1970s and 1980s (Drake, Essock, & Bond, 2009). During this period, many demonstration
projects produced disappointing results, in part due to poor model specification (Brekke, 1988). The
absence of clear, objective program standards—criteria for implementing a practice—interfered with
many well-intentioned efforts to disseminate these practices and severely attenuated the
accumulation of scientific evidence in support of these program models. In the MHTS, the target
audiences include the Social Security Administration, state and federal policymakers, advocacy
groups, the research community, Social Security beneficiaries, and the public at large. The report of
findings will better specify the intervention provided and will make it possible to replicate with use
of a reliable and valid fidelity scale in a multisite randomized controlled trial.
A second reason for assessing fidelity is to document variation across sites in a multisite study. The
MHTS was the largest demonstration ever of the IPS model, conducted in 23 study sites.
Understanding the level of variation in the context of the study outcomes provides policy guidance
for future implementations. The MHTS will provide information on the generalizability of the
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-2
overall intervention model across sites. One question is how feasible is it to offer the IPS model
with high fidelity?
The third reason to assess fidelity is to improve implementation. If collected periodically throughout an
intervention, fidelity assessments can provide feedback regarding changes needed to ensure full
implementation of the evidence-based practice. In other words, how feasible is it to modify
programs when they are not operating at high fidelity (Bond, Drake, McHugo, Rapp, & Whitley,
2009) and to sustain those practices over time (Rapp, Goscha, & Carlson, 2010)? If it is possible to
make program improvements through fidelity reviews, then this will greatly enhance the quality of
the research and future adaptation.
A fourth use of a fidelity measure is to predict outcomes. The general issue is measurement of the extent
to which adherence to the program model is associated with better outcomes. Program-level fidelity
has been found in prior research to be a strong and consistent predictor of client outcomes (Bond,
Becker, & Drake, 2011).
Receipt of IPS and OBH and related services is important for similar reasons as outlined for fidelity.
To facilitate communication, it is essential to describe services provided. Documenting variation
across sites offers guidance in assessing the degree of generalizability. Additionally, receipt of
services is also hypothesized to be associated positively with client outcomes (Bond & Kukla, 2011;
Ryan, Sherman, & Bogart, 1997).
MHTS investigators hypothesized that receiving integrated services (assistance for different needs
provided by a single treatment team) rather than fragmented services (e.g., assistance provided by
different service agencies or multiple sources) would be more effective, and as such, positively
influence the study outcomes. Therefore, it was important to not only examine whether services
were provided, but also to examine how services were provided (Corrigan, Mueser, Bond, Drake, &
Solomon, 2008). This general finding is well-established for the integration of behavioral health and
employment services (Cook et al., 2005; Drake, Becker, Bond, & Mueser, 2003) and for psychiatric
and substance abuse treatment (Drake & Bond, 2010). In the MHTS, the location of service (on-site
vs. off-site) served as one proxy for the degree of integration. However, we know that providing all
services at a single site does not guarantee integration of services, and investigators found that some
service systems were able to integrate across settings. The IPS Fidelity Scale also contains a global
measure of integration of employment services with mental health treatment based on interviews
with staff, observation of meetings, and medical charts. While this item is not an ideal measure of
integration, it is the best available measure of integration obtained in the MHTS. A more ideal
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-3
measure of integration would be a psychometrically-validated, comprehensive, multi-item scale using
multiple data sources. Such a scale has not yet been developed.
The remaining sections in this chapter describe the methods for assessing implementation, data
collection procedures, and the results.
Methods
Overview of Treatment Intervention
The treatment intervention had four service components: (1) IPS SE, (2) OBH and related services,
(3) SMM, and (4) treatment coordination by the NCC. The topics covered in this chapter include an
examination of implementation of IPS SE, and OBH and related services giving emphasis to the
quality of service delivery, timeliness of service delivery, service delivery rates, and service
integration. Additional topics include a description of the site variation in IPS implementation and in
the provision of the various elements of other behavioral services. The chapter concludes with an
examination of the relationship between IPS fidelity and site-level employment rates. A review of
the implementation of the remaining intervention components, SMM and the NCC role, follows in
Chapter 6. Treatment group participants also received financial support for insurance premiums and
out-of-pocket costs. Chapter 7 presents a detailed review of the financial support provided to
treatment group participants.
SE. Chapter 1 of this report describes the literature on the IPS model of SE. In sum, IPS is a welldefined form of SE and is an evidence-based practice specifically designed to serve individuals with
severe mental illness (Becker & Drake, 2003).
OBH and related services. The service needs of people with severe mental illness have been
identified in the literature (Corrigan et al., 2008; Drake et al., 2001; Kreyenbuhl, Buchanan,
Dickerson, & Dixon, 2010). In addition to IPS and SMM, the investigators identified the following
eight service categories, which they hypothesized to be essential services to which SSDI beneficiaries
should have access:

General medical care;

Mental health case management;
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-4

Social skills training;

Financial assistance;

Housing assistance;

Substance abuse treatment;

Family counseling; and

Legal assistance.
Throughout the study, MHTS investigators tracked the extent to which treatment group participants
received these services. Unfortunately, the investigators were not able to assess the intensity or
quality of implementation.
MHTS implementation of the IPS model. To help ensure fidelity to the IPS model, the
investigators used several strategies. First, as described in the section titled Study Site Selection in
Chapter 2, investigators purposively selected sites based on their historical commitment to
implementing high-fidelity IPS services. Specifically, a majority of the study sites were part of a
learning collaborative, which had been developed as a component of the Johnson & Johnson-Dartmouth
Community Mental Health Program (Becker et al., 2011; Drake, Becker, Goldman, & Martinez, 2006).
Second, three consultants were employed to serve as Quality Management Project Directors
(QMPDs). The QMPDs were experienced clinicians and program leaders who were experts in IPS.
The QMPD role was that of a trainer-consultant to the sites regarding implementation of the
psychosocial components of the intervention, with special attention to IPS implementation. The
three QMPDs shared the work of monitoring the 23 study sites in the MHTS. Each monitored
approximately one-third of the sites for the entire study period. The QMPD role was modeled on
the trainer role in a national multisite implementation study (McHugo et al., 2007).
The QMPDs made weekly telephone contact with the NCC and periodically spoke with the
program director at the site and IPS team leader. The primary focus of these calls was to review the
progress of each participant and help the NCC develop good IPS and clinical intervention plans.
Consultations also included discussions about the research protocol (e.g., recordkeeping and form
completion); assessments and overall workflow to implement the project; and how the NCC role
related to the clinical team, IPS specialists, and implementation of the IPS model to ensure fidelity.
Another focus of consultation was on establishing an alliance between clinicians and researchers.
QMPDs fielded questions about access to services, service needs of the study participants, care
coordination, and emergency treatment interventions. Also addressed were questions regarding the
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-5
authority of the NCC within the organizational structure of the study site. QMPDs noted that for
nurses, functioning in the capacity of a coordinator was typically not a funded position within the
community mental health center organizational structure. Many study sites hired the NCCs
specifically for conducting the unique job function and role specific to the MHTS. As a result, most
of the NCCs had limited experience working on a mental health team that provided IPS or
employment services. The QMPDs made sure to address employment-related updates on all calls
(i.e., any new jobs obtained, follow-along supports provided, barriers to gaining employment, and
reasons for job terminations).
The QMPDs also made annual site visits and conducted fidelity assessments of the IPS program,
using the IPS Fidelity Scale (Bond et al., 2011; Bond, Becker, Drake, & Vogler, 1997) to provide site
directors recommendations for improving IPS implementation (Bond et al., 2009) at their respective
site. Thus, the expectation was that all 23 MHTS study sites would achieve high fidelity.
Data Collection Procedures for Assessing MHTS Implementation
Data collection used to assess the implementation of the SE services and OBH and related services
of the treatment intervention required multiple strategies. First, to assess IPS fidelity at the programlevel, QMPDs conducted daylong annual site visits and systematically coded their observations using
the IPS Fidelity Scale (Appendix 5A). Second, to assess beneficiary-level receipt of services, NCCs
completed electronic service utilization reviews using the SE/OBH Quality Management Template.
Third, to obtain a global view of site implementation from the perspective of the QMPD, the
QMPDs completed a QMPD Site Report on two occasions, once in 2008 and once in 2009.
Discussions of these strategies appear in detail throughout the remainder of this chapter.
Measurement of IPS Fidelity at Program-Level
The quality of IPS implementation was assessed at the program-level using a 15-item measure
known as the IPS Fidelity Scale (Bond et al., in press; Bond et al., 1997) (Appendix 5A). The format
and assessment procedures for the IPS Fidelity Scale follow the conventions formalized in the
National Implementing Evidence-Based Practice Project (McHugo et al., 2007). Each scale item
reflects a specific element in the practice. A 5-point behaviorally anchored scale provides the rating
range. A rating of “5” indicates close adherence to the model, while a rating of “1” represents a
substantial lack of model adherence. For example, a score of five (5) for rapid job search indicates that
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-6
a first contact with an employer is, on average, within one month after program entry, whereas a
score of one (1) represents a delay of up to one year after program entry. Ratings of four (4), three
(3), and two (2) represent gradations between these two extremes.
For quality improvement purposes, the QMPDs provided feedback to sites about their relative
attainment of core elements in the IPS model using the item-level fidelity ratings. In addition, the
average of the item ratings yielded a total fidelity score, which expressed a global picture of overall
program-level fidelity. Thus, the total fidelity score ranged from one (1) to five (5), with higher
scores indicating more faithful implementation. Bond et al. (1997) established the following
benchmarks: a cutoff score of 4.33 (total score > 65) represents high fidelity, while 3.66 (total score
> 55) (but less than 4.33) represents fair fidelity. Although these benchmarks were originally based
solely on clinical judgment, these conventions have served the vocational rehabilitation field well by
communicating objectively the level of attainment of fidelity. IPS fidelity ratings are used to define
higher reimbursement rates for employment services in three state systems (Bond et al., 2011). The
utility of the IPS fidelity standards has been examined in 10 studies conducted by different research
groups from around the world, though refinement of the IPS fidelity standards continues (Bond et
al., 2011).
IPS Fidelity Scale assessment procedures. Typically, two trained fidelity assessors assessed the
IPS Fidelity Scale over the course of a one- to two-day site visit. One assessor was the QMPD
assigned to the site and the other assessor was a member of the MHTS investigative team. Because
the MHTS IPS services in many sites were a subcomponent of a larger IPS team, and IPS services
directed to MHTS treatment participants sometimes varied from the usual IPS services, the
assessors rated IPS services as delivered to MHTS treatment group participants.
In normal IPS fidelity assessments, assessors follow a detailed protocol with instructions for
preparing sites for the visit, critical elements in the fidelity assessment, and sample interview
questions (Becker, Swanson, Bond, & Merrens, 2008). For the MHTS, the assessors typically
interviewed the employment program leader and two or more employment specialists; observed
clinical team meetings where the NCC and the IPS specialist participated; shadowed employment
specialists on community contacts with employers; interviewed treatment group participants; and
reviewed beneficiary charts. When two assessors participated in an IPS review, each assessor
independently made fidelity ratings. The two assessors then reconciled any discrepancies to arrive at
a final set of fidelity ratings. For quality improvement purposes, the QMPDs prepared a fidelity
report summarizing the fidelity ratings and provided recommendations concerning any components
of the program that were deficient.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-7
Psychometric properties of the IPS Fidelity Scale. The IPS Fidelity Scale has excellent interrater
reliability (range of .67 to .99 for individual items and .98 for the total scale) and adequate internal
consistency (Cronbach’s alpha = .92). Nine of 10 studies assessing its predictive validity found
positive associations with employment outcomes. Its use in quality improvement has been
supported by positive reports from seven multisite projects (Bond et al., 2011).
In addition to measuring total fidelity, the scale also examines item-level fidelity. One item of
particular interest was integration of rehabilitation with mental health treatment, which often is the most
difficult component of IPS fidelity to achieve (Campbell et al., 2007).
SE/OBH Quality Management (QM) Templates
The Supported Employment/Other Behavioral Health (SE/OBH) QM template (form) was
designed for the MHTS and implemented at startup. Prior to project startup, MHTS investigators
developed a template that the NCC would complete on each treatment group participant. The intent
was to describe and track services received during the course of the study. For each participant, the
NCC completed a summary report covering the following time intervals: 0-3 months, 3-6 months, 612 months, 12-18 months, and 18-24 months. As part of the report, the template provided the
format for assessing elements in the IPS model as well as receipt of OBH and related services.
Systems staff programmed the template into the electronic Study Management System (SMS),
including automated reminders that triggered the NCC to complete the form at regularly scheduled
intervals.
The NCCs received training in the completion of the template at the project kickoff in June 2006.
Subsequently, the QMPDs provided consultation to the NCCs regarding questions that surfaced
about the template completion.
Based on the investigators’ review of the early template data, concerns voiced by the NCCs about
the template’s redundancy, and the QMPDs assessment of informational shortcomings of the tool, it
was determined that the template was not adequately eliciting quantifiable and reliable information.
The original template design failed to capture some of the critical data elements needed by the
investigators (for example, whether services were being delivered on-site or off-site). Consequently,
the investigators met in November 2007 to tailor and simplify the procedures for completion of the
template and to capture only the most salient items. The Results section of this chapter report the
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-8
findings for the common items on the original and revised templates, as well as the new items
developed for the revised template.
The IPS data elements assessed in the revised template included the following:

Employment status (was participant employed during period, and if yes, start and end
dates);

Engagement in MHTS services;

Vocational profile completed or updated;

Vocational plan completed;

Benefits counseling received;

Number of employer contacts;

Number of job interviews;

Number of calendar months in which there was a contact with the employment
specialist; and

Number of calendar months in which the employment specialist made contact with the
employer.
OBH and related service elements assessed on the template included whether a service was needed,
whether the participant received it, and (after the template revision described above) whether it was
received on-site or off-site. Overall, NCCs completed 95 percent of the SE/OBH QM templates
with site completion rates ranging from 75 percent to 100 percent.
Issues affecting analysis of SE/OBH QM template. The SE/OBH QM templates incorporated
the participant engagement question well after the study was underway. Consequently, a sufficient
sample to report on engagement for the first two reporting periods (0-3 months and 3-6
months) was not available, and therefore is not included in this report. In addition, three of the IPS
variables (vocational profile completed or updated, vocational plan completed, benefits counseling
received) were aggregated across the entire 2-year reporting period. The investigators concluded that
it was sufficient to count achievement of these milestones. Because initial employer contacts and job
interviews apply primarily to people who are not working, these analyses were limited to the not
employed subgroup during each reporting period. Most not employed treatment group participants
had no employer contacts or interviews, or if they did, by far the modal rate was a single contact.
Because of the low frequency of employer contacts and job interviews for each reporting period, the
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-9
investigators decided to dichotomize these variables into yes or no responses. The subgroup
examined in these analyses was further limited by excluding treatment group participants who the
NCC rated as unengaged during the reporting period.
In the OBH section of the template, inspection of the pattern of responses to the items concerning
need for services indicated virtually no difference between need and receipt of service. The MHTS
investigators concluded that this item might have been difficult for the NCCs to assess and that the
responses may not be valid for analysis. Specifically, the data suggested that the NCCs were inferring
need from the receipt of services, essentially obviating the distinction between need and receipt. Thus,
there was no reliable assessment of need for services and the investigators dropped from the analysis
“need for service” items included in the SE/OBH QM template.
The SE/OBH QM template also included an item regarding receipt of medication management
services. Careful inspection of this item strongly suggested that some NCCs interpreted this question
to be indicative of the need for a participant to receive training, intervention, or assistance with
taking his or her prescriptive medicines rather than an affirmation of the receipt of a psychotropic
medication. Therefore, the investigators dropped this item from this set of analyses. Instead, the
investigators assessed medication management through the Systematic Medication Management
Quality Management reporting system. That data is included in Chapter 6 of this report. Finally,
MHTS investigators studied the temporal patterns for receipt of OBH and related services. The
overall rates of receipt of services within each service category were highly stable. Thus, this chapter
reports only the mean rates of receipt of OBH services across all reporting periods for the 2-year
time period. The original OBH template did not collect on-site or off-site data; therefore, this report
contains the mean on-site and off-site rates for the last two reporting periods only.
QMPD Site Reports on Site Implementation
In January 2008 and again in January 2009 the three QMPDs completed a global report on each of
their study sites for the preceding calendar year using a structured checklist labeled the QMPD Site
Report. This report supplemented information captured in the IPS Fidelity Scale, adding more
detailed description of site level activities and characteristics specific to the implementation of the
MHTS. These observations provided additional explanations for site-level performances not
captured in the IPS Fidelity Scale.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-10
The checklist contained 73 items consisting of yes or no responses, Likert scale ratings, and
estimates of percentages. A comments section for each item permitted the QMPD to provide
qualitative responses. The checklist included nine domains, each designed to assess various aspects
regarding the quality of implementation of MHTS services: Global Rating of Implementation of IPS
Services, Steering Committee and Leadership, Other Behavioral Services, Study Site Director, SE
Leader, SE Staffing, Mental Health Treatment Team, NCC Services, and Financing.
Results
Program-Level Fidelity to the IPS Model
Based on the 15-item IPS Fidelity Scale, 77 percent of study sites achieved high fidelity within the
first year (see Figure 5-1). An even higher proportion achieved high fidelity in the second and third
years (approximately 86% of sites in both years). Conversely, across 65 fidelity reviews, the fidelity
assessors rated one active site as having poor fidelity in any annual assessment. Overall, IPS fidelity
ratings averaged 67.6 in Year 1, 69.3 in Year 2, and 67.2 in Year 3 (the total possible score is 75).
Because of the high level of attainment of fidelity overall, most IPS fidelity items showed little
variability. However, the item on Integration of IPS with Behavioral Treatment was generally a more
difficult item on which to attain high fidelity. As shown in Figure 5-2, 55 percent of the sites
achieved the highest score for full integration in Year 1. By Year 3, two-thirds of the study sites had
achieved full integration.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-11
Figure 5-1.
IPS fidelity for MHTS sites (based on 15-item IPS Fidelity Scale)
100.0%
4.8%
13.6%
22.7%
9.5%
80.0%
60.0%
Low
40.0%
86.4%
77.3%
85.7%
Fair
High
20.0%
0.0%
Year 1
(N=22)
Year 2
(N=22)
Year 3
(N=21)
NOTE: Years 1 and 2 did not include the study site added in the third year of the study. Year 3 did not include sites that completed study
participation (i.e., all treatment group participants transitioned from the study) at the time of the fidelity visit.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-12
Figure 5-2.
Site integration of IPS and behavioral treatment (IPS Fidelity Scale item)
100%
18%
32%
24%
80%
23%
60%
10%
14%
Poor
Moderate
40%
55%
59%
67%
Full
20%
0%
Year 1
(N=22)
Year 2
(N=22)
Year 3
(N=21)
Engagement in IPS
Analysis of participant engagement in IPS is an important aspect that is part of the fabric of
implementation. NCCs rated engagement on the QM template based on their review of participant
records and consultation with the participant and the SE specialist. In this descriptive analysis, the
investigators assumed engagement by employed treatment group participants (during a reporting
period). Thus, the focus in this analysis was on treatment group participants who were not employed
during the specified reporting period (see Table 5-1). According to the SE/OBH QM templates, the
unengagement rate was 8 percent in the 12-18 month reporting period and 11 percent in the 18-24
month reporting period. During the 12-18 month reporting period, site-level unengagement rates
(not shown) ranged from 0 percent to 19 percent; three sites had unengagement rates of 15 percent or
higher, while nine sites had unengagement rates of 5 percent or less. During the 18-24 month
reporting period, site-level unengagement rates (not shown) ranged from 0 percent to 30 percent; nine
sites had unengagement rates of 15 percent or higher, while eight sites had unengagement rates of 5
percent or less. The engagement rate for treatment group participants is much higher at both 18
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-13
months and 24 months than found in many studies of mental health treatment. For example, a
review by Kreyenbuhl, Nosse, and Dixon (2009) found that “up to one-third of individuals with
serious mental illnesses who have had some contact with the mental health service system disengage
from care” (p. 696).
Table 5-1.
IPS service engagement rates for employed and not employed treatment group
participants in the last two reporting periods
Beneficiary group
Engaged and employed
Engaged and not employed
Not engaged and not employed
Missing2
1
12 to 18 months
(N=9811)
freq
%
356
36.3
524
53.4
81
8.3
20
2.0
18 to 24 months
(N=981)
freq
%
354
36.1
478
48.7
111
11.3
38
3.9
This number reflects the total number of treatment group participants who completed 24 months in the MHTS (i.e., excludes
administrative drops, withdrawals, and deaths) with the exception of one beneficiary who was miscoded at the time of these analyses.
2 The
missing category includes participants for whom the NCC did not complete the QM SE template.
Beneficiary-Level Receipt of IPS
Three important elements of IPS services are the provision of benefits counseling, conducting a
vocational assessment, and formulating a vocational plan. As shown in Table 5-2, 69 percent of the
treatment group participants received benefits counseling at some point during the 2-year study
period, with wide variation across sites (9% – 98%). Eleven sites (48%) provided benefits counseling
to 80 percent or more of the treatment group participants, nine sites (39%) provided benefits
counseling to 46 percent to 74 percent of participants, and three sites (13%) provided benefits
counseling to 31 percent or fewer of participants. In reporting these statistics, it should be noted
that receipt of benefits counseling at any time is a modest criterion of implementation; optimally
treatment group participants would receive benefits counseling whenever their benefits are expected
to be affected (e.g., when contemplating starting or changing jobs).
Table 5-2.
Percentage of treatment group participants who received IPS service elements
while enrolled in the MHTS
IPS elements
Benefits counseling
% Received service at any time
69
Site-level % (Range)
9% - 98%
Vocational profile
90
61% - 100%
Vocational plan
94
63% -100%
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-14
As shown in Table 5-2, regarding both the vocational assessments and vocational plans, the overall
completion rates exceeded 90 percent, but there was site variation on these indicators as well, with
some study site rates below 65 percent. The timing of completing the assessment and vocational
plans have not been reported here, although in a high-fidelity IPS program, IPS clients should
complete these milestones within the first 3 months. Early completion of these milestones was the
modal pattern for treatment group participants in the MHTS.
Table 5-3 shows the rates of contact across reporting periods among not employed treatment group
participants, excluding the unengaged subgroup. Treatment group participants could meet multiple
times with employment specialists. This table reports the mean number of months per beneficiary
when there was at least one contact with an employment specialist. During the first reporting period
(months 0-3), not employed treatment group participants met with their employment specialist on
average 2.2 months out of the three, or 73 percent of the available months. This rate declined for
each subsequent reporting period, dropping to a rate of 53 percent in the 18-24 month reporting
period. The criterion of one contact per month is a very modest criterion; in high-fidelity programs
clients looking for work would be expected to meet weekly or even more often (Leff et al., 2005).
As shown in Table 5-3, the rate of contact with employers, by treatment group participants who
were not employed during a reporting period or by their employment specialist, ranged from 25
percent to 40 percent across the 5 reporting periods, with the highest rate during the period from 6
to 12 months. Mirroring these rates was the percentage of treatment group participants with at least
one.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-15
C
S
Chapter
5: Implementation off Supported
Employment and Other Behavioral Health and
Related Services
Table 5-3.
IPS employment service contact rates of engaged but not employed treatment group participants during the reporting
period
Months 0-3
3 months
Months 3-6
3 months
Months 6-12
6 months
Months 12-18
6 months
Months 18-24
6 months
Mean of all
reporting periods
4 months
Mean # months with at
least one IPS contact
during interval(s)
2.2 (73%)1
2.1 (70%)
3.7 (62%)
3.4 (57%)
3.2 (53%)
63%
N (%) with at least one
contact with employer
83 (25%)
102 (28%)
178 (40%)
167 (32%)
142 (31%)
31%
4 sites ≥ 50%
3 sites ≤ 10%
3 sites ≥ 50%
5 sites ≤ 10%
7 sites ≥ 50%
3 sites ≤ 10%
7 sites ≥ 50%
3 sites ≤ 10%
5 sites ≥ 50%
6 sites ≤ 10%
3 sites ≥ 50%
3 sites ≤ 10%
65 (19%)
72 (20%)
119 (27%)
100 (19%)
76 (17%)
20%
3 sites ≥ 50%
5 sites ≤ 10%
1 sites ≥ 50%
5 sites ≤ 10%
3 sites ≥ 50%
3 sites ≤ 10%
1 sites ≥ 50%
4 sites ≤ 10%
0 sites ≥ 50%
9 sites ≤ 10%
0 sites ≥ 50%
4 sites ≤ 10%
335
359
443
519
455
Interval
Variation across sites
on employer contacts
5-16
N (%) with at least one job
interview
Variation across sites
on job interviews
Total N2
1
2
Percentage is expressed in person-months.
The total N excludes treatment group participants with missing QM templates for the given reporting period. In addition, this table does not include treatment group participants identified
as unengaged (no contacts and not employed) during each of the last two reporting periods (see Table 5-1).
job interview during a reporting period, ranging from 17 percent to 27 percent. Using greater than
or equal to 50 percent as the criterion for a high-performing site, the number of high-performing
sites ranged from three to seven sites on contacts with employers and ranged from zero to three
sites on job interviews across reporting periods. Using the criterion of less than or equal to 10
percent as the criterion for a low-performing site, the number of low-performing sites ranged from
three to six sites on contacts with employers and ranged from three to nine sites on job interviews
across reporting periods.
The NCCs reported much higher rates of monthly contact among employed treatment group
participants, compared to the rates for not employed treatment group participants, as shown in
Table 5-4. Overall, employed treatment group participants met with their employment specialist in
75 percent of the beneficiary months across their 24 months of participation in the study, with the
rates ranging from 68 percent to 85 percent across reporting periods. The rates of followup contact
with employed treatment group participants declined slightly over time. This decline is consistent
with the IPS literature (Bond & Kukla, 2011).
Sites varied considerably in rates of followup by employment specialists with treatment group
participants. During the first 6 months, 11 to 14 sites had consistent followup with employed
treatment group participants (defined as 5 of 6 months with contact). The number of sites with
consistent followup with employed treatment group participants tapered off to three sites during the
last three reporting periods. Conversely, the number of sites with infrequent contact with employed
treatment group participants (defined as contact in one-half or less of the months in the reporting
period) increased from one to four sites over the period of participation.
Table 5-4.
Rates of IPS contact with employed treatment group participants by reporting
period
Mean # months with an
IPS contact
Variation across sites
on IPS contacts
Total N2
1
2
Months 0-3
Months 3-6
Months
6-12
Months
12-18
Months
18-24
2.6 (85%)1
2.5 (83%)
4.3 (72%)
4.1 (68%)
4.1 (69%)
14 sites ≥ 2.5
1 site ≤ 1.5
11 sites ≥ 2.5
1 sites ≤ 1.5
6 sites ≥ 5
2 sites ≤ 3
4 sites ≥ 5
4 sites ≤ 3
3 sites ≥ 5
4 sites ≤ 3
87
164
319
353
347
Percentage is expressed in person-months.
The total N excludes treatment group participants with missing QM templates for the given reporting period.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-17
Mean of
reporting
periods
75%
Receipt of OBH and Related Services
Rates of receipt of OBH and related services were examined for all treatment group participants
who completed 24 months in the study (N=981). Table 5-5 shows a summary of those data. The
reported rates are averages across the five reporting periods, which were generally stable. Overall,
the study sites provided more than half of the treatment group participants with case management
services. This rate does not include any case management services provided by the NCCs. Similarly,
53 percent of treatment group participants received general medical care. The rate of receipt of the
other 6 behavioral services ranged from 7 percent to 21 percent.
Table 5-5.
Percentage of treatment group participants who received OBH or related services
while enrolled in the MHTS1
Services
Mental health case management
General medical care
Social skills training
Financial assistance
Housing assistance
Substance abuse treatment
Family counseling
Legal assistance
1
2
% of treatment group
participants
who received specific
behavioral service
54
53
21
16
15
13
8
7
% of treatment group
participants receiving service
who received specific
behavioral service on-site2
72
12
68
68
69
44
64
37
Rate of receipt of services based on an average across all five reporting periods. The sample size for each reporting period varied
based on the number of completed QM templates. The mean sample size for five reporting periods was 732. The denominator for
calculating percentages includes unengaged treatment group participants.
Rates of receipt of services on-site were based on an average of the last three reporting periods because the question was only
included in the revised template.
Table 5-5 also shows the percent of treatment group participants who received OBH or related
services, and the percentage of that group who received the service on-site at the study site, as
opposed to off-site at another location with a different service provider. One might expect general
medical care to be primarily an off-site service, with the remaining seven services offered on-site.
The data revealed that this was not consistently the case, as rates for these other on-site services
ranged from a low of 37 percent to a high of 72 percent.
MHTS investigators examined more carefully the site-level variation in the receipt of OBH and
related services. For each behavioral service category, sites were classified as “high access,”
“moderate access,” or “low access,” based on the percentage of treatment group participants
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-18
receiving each specific service. The criteria for high and low access were based on the distribution of
rates of services within the MHTS sample of 23 sites. The summary overview of site variation in the
rate of high and low access sites is shown in Table 5-6.
Table 5-6.
Frequency and percentage of high and low access sites for other behavioral health
services
Other Behavioral Service
Mental health case management
Social skills training
Financial assistance
Housing assistance
Substance abuse treatment
Family counseling
Legal assistance
High Access Sites
Criterion for
high access
freq
%
≥80%
6
26.1
≥70%
1
4.3
≥30%
5
21.7
≥33%
3
13.0
≥33%
1
4.3
≥20%
3
13.0
≥10%
5
21.7
Low Access Sites
Criterion for
low access
freq
%
≤25%
6
26.1
≤10%
9
39.1
≤10%
11
47.8
≤10%
8
34.8
≤10%
10
43.5
≤10%
15
65.2
≤10%
18
78.3
Site-level rates of substance abuse diagnosis and treatment varied across sites. Six sites had substance
abuse diagnosis rates of 50 percent or more (based on the Structured Clinical Interview for DSM-IV
Axis I Disorders [SCID]), while 5 sites had rates of 31 percent or less. In terms of receipt of
substance abuse treatment, one site reported that 33 percent of all its treatment group participants
received these services. No other site provided substance abuse treatment to more than 30 percent
of all treatment group participants. Ten sites reported less than 10 percent of all treatment group
participants received substance abuse treatment services. The correlation between site-level rates of
diagnosis of substance abuse and receipt of substance abuse treatment was .11 (non-significant). The
lack of correlation suggests that (1) need did not determine service provision and (2) integrated
substance abuse treatment was not offered by all the sites.
In 5 sites, 75 percent or more of treatment group participants receiving substance abuse treatment
received it on-site, while in 6 sites less than 10 percent received it on-site.
There was wide variation in the provision of case management services. At six sites 80 percent or
more of treatment group participants received case management, while at six other sites 25 percent
or fewer of treatment group participants received case management. Most sites provided case
management on-site. In 13 sites, 90 percent or more of the treatment group participants receiving
case management received it on-site, while 4 sites provided case management on-site to less than 10
percent of treatment group participants who received case management.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-19
One site provided social skills training to 70 percent of treatment group participants. No other site
provided social skills training to more than half of treatment group participants, while 9 sites
provided social skills training to fewer than 10 percent of treatment group participants. When it was
provided, social skills training was typically on-site, with 11 sites providing it to three-fourths or
more of treatment group participants who received it.
Few sites provided extensive housing services, which was likely due in part to limited need among
treatment group participants. Three sites provided housing services to more than one-third of their
treatment group participants, while 8 sites provided housing to 10 percent or fewer of treatment
group participants. When provided, housing services were most often on-site; 10 sites provided
housing on-site in 80 percent or more of cases.
Provision of family counseling was very rare, with 15 sites providing these services to less than 10
percent of treatment group participants. Only 3 sites provided families counseling to as many as 20
percent of treatment group participants, and for these 3 sites the family services were on-site.
Provision of financial assistance also varied across sites, with 5 sites providing help to 30 percent or
more of treatment group participants while 11 sites provided help to 10 percent or less. Financial
assistance typically was on-site; fourteen sites provided financial assistance on-site in 75 percent or
more of cases. Sites rarely offered legal assistance; only five sites provided assistance to 10 percent or
more, with 22 percent being the highest rate overall. Location of legal assistance was quite variable.
QMPD Site Reports
Table 5-7 shows the QMPD global ratings of IPS implementation. The results show that 86 percent
of sites had adequate-to-very strong implementation during 2008, and 77 percent had adequate-tovery strong implementation during 2009. In other words, there were three to five sites with
substandard implementation during these periods. These results roughly correspond to the IPS
fidelity ratings. As part of an exploratory analysis, the investigators asked the QMPDs to identify up
to three barriers to implementation. It should be understood that this was in the context of
understanding barriers in the context of a group of sites that generally had achieved the project goals
regarding IPS service implementation according to the quantitative data reported earlier in this
chapter. As shown in Table 5-8, the most commonly mentioned barriers to implementing the
treatment in the 2009 site report were unresponsive leadership, finances, and unavailability of mental
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-20
health services. Regarding the salience of leadership as a barrier to implementation, these findings
are consistent with other studies in the literature (e.g., Torrey et al., 2011).
Table 5-7.
Very strong
Strong
Adequate
Substandard
Table 5-8.
Frequency and percentage of MHTS study sites that received various overall quality
of site-level IPS implementation ratings based on the QMPD Site Reports: 2008 and
2009
Ratings
freq
9
4
7
3
2008
(N=23)
%
39.1
17.4
30.4
13.0
freq
5
6
6
5
2009
(N=23)
%
21.7
26.1
26.1
21.7
Frequency and percentage of MHTS study sites with various site-level barriers to
implementation based on QMPD qualitative assessments
Barriers
Leadership: unresponsive/not committed
Finances
Mental health services not available/not integrated
Staff turnover
Failure to follow model (e.g., time in community)
Lack of understanding or experience
Supervisor: lack of time, unresponsive to feedback
State mental health actions
Transportation
Offices outside the Mental Health Center
Vocational rehabilitation access
Poor communication
Local economy (unemployment rate)
Capacity: Not prepared to work with # clients
Outside agency provide staff
Poorly paid staff
Start-up issues
freq
10
7
6
5
5
4
4
3
2
2
1
1
1
1
1
1
1
%
45.5
31.8
27.3
22.7
22.7
18.2
18.2
13.6
9.1
9.1
4.5
4.5
4.5
4.5
4.5
4.5
4.5
NOTE: Assessments completed in January 2009 on all 23 MHTS study sites.
MHTS Site-Level Fidelity-Outcome Analysis
MHTS investigators also examined associations between site-level IPS fidelity and site-level
employment rates for the treatment group. The measure of IPS fidelity was the total IPS fidelity
score for each site, assessed annually, as described earlier in this chapter. The total site scores for
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-21
Years 1-3 and the mean across the three years served as predictors. The site-level employment rate
served as the employment outcome measure.
Employment rates were based on beneficiary self-reports (i.e., the responses on the computer
assisted personal interviews [CAPI], as described in Chapter 2), which indicate the percentage of
treatment group participants employed at any time during the two-year period, as reported in the
Quarterly and final Followup interviews (see Data Collection Procedures section of Chapter 2). As
shown in Table 5-9, IPS fidelity ratings were not associated with self-reported employment rates.
Table 5-9.
Site-level fidelity-outcome correlations in the MHTS
Fidelity
CAPI (self-report) Employment Rates
Site employment rate for treatment
Year 1
(n = 22)
Year 2
(n = 22)
Year 3
(n = 21)
Mean
Years 1-3
(n = 23)
-.10
.10
.05
-.02
Discussion
This analysis of the implementation of IPS and OBH and related services in the MHTS suggests
some areas of excellence as well as other areas falling short on the criteria for the planned model.
The most important finding was that 80 percent or more of sites achieved a high level of IPS
program implementation and maintained it across the entire study period. Given many instances in
the literature in which multisite trials have failed because of inadequate implementation (Bond, 2007;
Brekke, 1988; Drake et al., 2009; Michie, Fixsen, Grimshaw, & Eccles, 2009), the high level of
fidelity and documentation of that fidelity using a well-validated scale is a noteworthy achievement.
Even the National Implementing Evidence-Based Practices Project, which was designed to achieve
high fidelity using a comprehensive and standardized training-consultation strategy, achieved only a
55 percent success rate (McHugo et al., 2007). The success in the MHTS was due in part to the
judicious site selection along with careful monitoring and fidelity reviews provided by skilled
consultants (i.e., the QMPDs).
The review of beneficiary-level IPS fidelity was not as successful. Part of the challenge was lack of
beneficiary engagement with IPS services. The nominal rate of unengagement was relatively low—
less than 10 percent overall, but not employed treatment group participants appeared to have a
relatively low rate of employer contact. According to NCC reports, only a small fraction of not
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-22
employed treatment group participants received direct IPS program support for actively seeking
employment (in terms of employer contacts and job interviews) during the study period. What these
data do not show are the reasons for the lack of employer contact by the IPS program for employed
participants. The IPS model assigns responsibility for engagement to the IPS program (which
includes integrated behavioral health, such as case management) rather than attributing
unengagement to a lack of motivation on the part of IPS participants. In this sense, the IPS
programs fell short in employing effective engagement strategies. Investigators speculated that this
might have something to do with the particular clientele that included persons from the Social
Security Disability Insurance (SSDI) rolls who joined the study but had not been engaged in
treatment at any community mental health center. For example, the QMPDs had received reports
from NCCs regarding treatment group participants who found the population served at some of the
study sites to be too mentally disabled and unsettling to sit in the waiting room while waiting to meet
with staff.
With respect to other behavioral services, special attention was paid to receipt of mental health case
management services (i.e., assistance from mental health center staff to clients with regard to illness
management, practical issues such as shopping, cooking, and other activities of daily living, finding
housing, and obtaining food stamps and other entitlements), because case management is key for
achieving adequate mental health treatment (Rapp & Goscha, 2004). As is true in general throughout
the U.S., Medicaid typically was the primary source of funding for mental health case management
services to treatment group participants who received case management. The fact that NCCs
reported that only 54 percent of treatment group participants received mental health case
management is far below the expected rate in IPS programs serving clients with severe mental
illness. There are several mitigating factors, however. First, it may be assumed that some treatment
group participants did not need case management services. Unfortunately, as previously mentioned,
measuring need through QM reporting was not successful, thus negating the reliability of estimating
extent of need. Second, some treatment group participants received case management assistance
from the NCC. However, no formal data exist on the extent to which NCCs provided case
management services. Given their competing responsibilities, it is most likely that NCCs played only
a limited case management role—both in number of treatment group participants assisted and in the
range of case management activities performed. The QMPD site reports reinforce the conclusion
that at least at some sites, case management services were insufficient. Since case management is a
vital part of an integrated IPS team and was often unavailable, it may be the missing factor in
engaging treatment group participants to IPS. This may have been more of a factor in this study
than in typical IPS programs, because many of the treatment group participants were not engaged in
any mental health center services at the start of the project. In usual IPS practice, clients are typically
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-23
first engaged in mental health treatment and then referred to IPS. By contrast, in this study,
randomization into the treatment intervention occurred prior to the receipt of mental health
treatment.
Integration of IPS and behavioral health services was unevenly implemented. Converging evidence
from the IPS Fidelity Scale, provision of OBH and related services off-site, and QMPD site reports
suggest that integration was an ongoing issue for all sites, and an especially serious limitation in some
sites.
Based on the analyses reported in this chapter, the investigators conclude that it is possible to
develop and implement faithfully a program of treatment and SE and to engage reasonably large
populations of SSDI beneficiaries. Significantly, these services can be delivered not only to SSDI
beneficiaries who are enrolled in public mental health services prior to enrollment in IPS, but also
SSDI beneficiaries who have not been receiving mental health treatment in the public mental health
system. Many of the MHTS participants were not currently engaged in mental health services—
certainly not at the typical IPS clinical settings. Those who were engaged in mental health treatment
had a range of options for receiving behavioral health services other than their local community
mental health center, as most of them had Medicare as part of their SSDI benefits (and some had
Medicaid as well).
Because the investigators sought to maximize geographic dispersion in site selection and given that
the large majority of sites achieved high IPS fidelity, it is plausible to hypothesize that IPS is
transportable to throughout the U.S., though the study was not statistically powered or designed to
answer the question of generalizability across geographic regions. This study adds to the substantial
literature suggesting that IPS can be successfully implemented in different regions of the U.S.
(Becker et al., 2011).
While the IPS services at the program-level were available at high fidelity, job-seeking rates among
not employed treatment group participants were disappointing. IPS experts may need to make
modifications to the IPS model to address the characteristics of this population. This study has not
isolated the exact barriers to employment for this population (e.g., concern are about losing SSDI or
healthcare benefits, being less physically able to job search given the co-existence of physical
illnesses, etc.). Another factor in this study was the limited time horizon; in other words, one barrier
might have been the relative brevity of the followup period. This factor, however, was probably of
limited impact in that the probability of actively searching for competitive work declines sharply
over time for study participants in published studies of IPS (e.g., Bond et al., 2008 – PRJ review
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-24
paper). We conclude that one strategy for overcoming the barriers to employment is for IPS
programs, in conjunction with mental health treatment staff at mental health centers, to employ
intensive engagement efforts with this population.
This study found a surprisingly modest uptake of other behavioral services among treatment
participants. As is true in the general mental health services literature, the reasons for nonreceipt of
other behavioral services are complex and diverse (Drake & Essock, 2009). The current study was
not equipped to quantify the reasons for the unexpectedly modest uptake of other behavioral
services by a group of participants likely experiencing a range of untreated psychiatric problems.
One challenge in achieving precise estimates was the absence of psychometrically valid measures of
need for treatment. The investigators abandoned the assessment of need by NCCs when they
determined that their assessments were not credible. Therefore, one unknown factor was the extent
of need. Based on the findings from the mental health services literature (Drake & Essock, 2009),
another likely factor was participant refusal to accept treatment services. A third factor, which the
research team documented anecdotally in weekly consultation calls by the QMPDs to the NCCs and
through site visits, was the lack of access to services. Treatment group participants experienced
many instances of system and organizational barriers when in need of or receiving OBH services.
Drake and Essock (2009) found that the public mental health system is deeply underfunded and
access to services is difficult for anyone with mental health problems. Even with the provision of
full health insurance coverage, many treatment group participants did not receive the assistance they
needed because of limitations in access and supply of behavioral health providers as well as policies
and procedures at each site, which in turn were often prompted by state regulations.
Overall, investigators observed a wide variation across sites in provision of “related” services. State
policies, provider agency policies, agency culture regarding assertive engagement, leadership, and
clinical supervision accounted for the wide variance of site differences.
Finally, the investigators examined one predictor of employment outcomes based on site-level
variation in IPS fidelity. IPS fidelity was not associated with the beneficiary-reported employment
rate; this lack of correlation was more than likely a result of restriction of range. While there was
some variation among sites in fidelity, it may not have been sufficient to detect differences in
employment rates attributable to fidelity of program implementation.
Chapter 5: Implementation of Supported
Employment and Other Behavioral Health and
Related Services
5-25
Chapter 6
Implementation of the Nurse Care Coordinator Role and Systematic
Medication Management
The primary treatment intervention components of the Mental Health Treatment Study (MHTS)
offered access to systematic medication management (SMM), Individual Placement and Support
(IPS) supported employment (SE), other behavioral health (OBH) and related services, and Nurse
Care Coordinator (NCC) services. While both IPS SE services and SMM are widely recognized as
having a strong evidence base, investigators have never fully tested these treatment modalities on the
Social Security Disability Insurance (SSDI) population or across a large number of community
mental health centers. The Social Security Administration (SSA) wants to know if it is feasible to
offer these evidence-based services in community mental health centers across the United States,
and whether providing SSDI beneficiaries with access to these services improves employment,
health, and quality of life. Chapter 4 presented convincing evidence that the combination of services
known as the MHTS treatment intervention indeed improves employment, mental health, and
quality of life. Chapter 5 presented evidence of program-level and individual-level fidelity of the IPS
SE model tested as a primary component of the treatment intervention. The purpose of this chapter
is to address the extent of implementation of the SMM and NCC service components of the
treatment intervention. The report covers four topics pertinent to the implementation of SMM.
These topics include an overview of the psychiatric diagnoses and health conditions presented by
treatment group participants, beneficiary and prescriber levels of engagement in SMM, the NCC role
in the study, and a summary of the SMM implementation and its impact. Prior to addressing these
topics, it is important to provide a background for understanding SMM and its intended operation in
the study.
Background
SMM services in the MHTS used an NCC to facilitate and promote prescriber use of evidence-based
guidelines and recommendations for medication management of severe and persistent mental
illnesses. SMM has its roots in two different areas of clinical research into treatment of major
psychiatric disorders. First, a number of studies of medication use and chart documentation in
treatment of schizophrenia have identified widespread problems among providers with medication
selection, medication dosing, medication side effect recognition and treatment, recognition and
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-1
treatment of persistent symptoms, and documentation of symptoms, side effects, and treatment.
Second, a large literature on recognition and treatment of depression in primary care settings has
demonstrated the utility of a team approach to the illness, in which a physician extender (typically a
registered nurse) functions to evaluate symptoms and provide relevant assessment documentation to
the treating physician with regard to need for treatment and medication selection. Given these
observations, the Technical Advisory Panel and SSA decided that inclusion of SMM in the MHTS
treatment intervention could contribute to improved outcomes by decreasing the deleterious effects
of inadequately treated illness symptoms and medication side effects.
An initial effort to devise a medication management program to remedy these identified deficiencies
was the Texas Medication Algorithm Project (TMAP), which developed manuals to guide treatment
of depression, bipolar disorder, and schizophrenia. Additionally, the manuals provided clinical
coordinators the ability to work with physicians in evaluating patients, providing patient education,
and in medication selection (Rush et al., 2003). A key component of TMAP was use of standardized
assessments and documentation to quantify illness symptoms, explicit identification of medication
side effects, and documentation of these observations so as to be accessible to present and future
providers.
Subsequent to completion of TMAP, there were a number of efforts in public mental health clinics
and systems to implement approaches to medication management based on TMAP. Observations of
successes and failures in these efforts supported two conclusions. First, the organization must take
the lead in SMM implementation, because it involves changes in medical recordkeeping, patient flow
through the system, and training of clinical teams. Second, the critical role of the clinical coordinator
as the “glue” that holds together the SMM approach was emphasized by the poor record of
implementation in sites that did not provide personnel to fill this role. Simply adding new tasks to
already overburdened prescribers usually generated resistance and poor implementation. A corollary
observation from TMAP and related projects has been that nurses have the best training to work
with prescribers on medication management issues, and that their credibility with prescribers and
patients in the medication arena is typically higher than for other non-physician mental health
professionals.
In the design of the MHTS, the investigators decided to hire an NCC at each study site. The NCC’s
primary roles were: (1) to implement the SMM program (described more fully on the next page);
(2) to monitor implementation of SE and OBH and related services (e.g., substance abuse treatment,
benefits or family counseling) when prompted; and (3) to promote integration of care among
participants’ mental and physical health providers. The NCC documented (on the SMM intake form
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-2
and NCC Clinical Reports) physical health conditions and whatever medication treatments the
treatment group participants were receiving from their providers, and communicated with the
physical health care provider if clinically indicated. The NCC also kept the mental health treatment
team informed about participants’ physical health conditions and their treatment. The potential
impact of physical illnesses and the treatments for them on functionality is clear, but it is also
notable that patients with major mental illnesses, as a group, have significantly earlier mortality from
medical conditions, with average life spans up to 25 years shorter than the general population (Parks
et al., 2006). The majority of this excess mortality is attributable to medical conditions such as
cardiovascular, pulmonary, and infectious diseases.
Two realities shaped the implementation of SMM in the MHTS. First, it was impractical for
participating sites to change their entire operational systems to accommodate implementation of
SMM. Therefore, a system for implementation of SMM just for treatment group participants was
developed, which depended heavily on forms and electronic data entry systems (i.e., the Study
Management System or SMS) created by the prime contractor. That is, only a subset of patients
received SMM, and clinic prescribers otherwise continued to function in their usual roles and duties.
This meant that prescribers used SMM for one group of patients, but not for others, which likely
affected the degree to which they bought into and adopted the SMM program as their modus operandi.
Second, since beneficiary recruitment for the study was from the community at large, it was
expected, and ultimately confirmed, that many treatment group participants would have ongoing
relationships with prescribers who had no connection with the clinical system within the study site in
which the NCC was working, and which was providing SE services. The MHTS investigators
considered it quite likely that “outside” prescribers would not fully participate in the SMM program
for both logistical and systemic reasons. However, the investigators concluded that it would be
unwise and unnecessarily exclusionary to require that participating beneficiaries receive their
medication management at the study site. Some participants opted to switch prescribers on their
own, but many chose to continue with their original off-site prescribers. The NCCs tried to work
with the prescribers at a distance, through e-mail, fax, and telephone contacts. At some sites, NCCs
made office visits to outside prescribers to introduce themselves and the SMM program. As will
become evident from the data presented later in this chapter, these accommodations to the realities
of participant prescriber selection and of existing site medication management systems had large
effects on implementation of SMM.
Figures 6-1, 6-2, and 6-3 depict decision-making strategies and interactions related to medication
selection. Figure 6-1 illustrates the contextual variables that affect prescriber medication decisions,
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-3
organized by those variables that are associated with the patient, those associated with the
medication, and those that are associated with the system within which the decisions are occurring.
Figure 6-1.
Factors influencing prescriber medication decisions
PATIENT LEVEL
• Medication history
• Psychiatric history
• Current symptoms/side effects
• Adherence/Non-adherence
• Concurrent physical illness
• Age, race, ethnicity
• Preferences
MEDICATION LEVEL
• Efficacy
• Tolerability
• Drug - drug
interactions
• Drug metabolism
• Dosing
SYSTEM LEVEL
• Formulary
availability/restrictions
• Cost to patient
• Cost to 3rd party payors
• Access/convenience issues
Prescriber
Medication
Decisions
Figure 6-2.
Factors influencing prescriber medication decisions in SMM
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-4
Figure 6-3.
Role-based functions in SMM
The central goal of SMM is to provide prescribers with key, updated information about the patient
and medication options at the time of each medication visit with the prescriber. Figure 6-2 depicts
the role of the NCC as an intermediary who assimilates relevant information and recommendations
and conveys them to the prescriber.
Figure 6-3 lists the inputs into the process used by the NCC in the MHTS, and shows the flow of
information from prescriber to NCC concerning actual medication decisions and the rationales for
them.
In the ideal SMM model, NCC and prescriber function in tandem, with the NCC first assessing the
patient’s status with regard to medication use, medication effects, and medication-related symptom
outcomes. The NCC conveys this information, along with any recommendations for change based
on medication management guidelines and expert consultation, to the prescriber. The prescriber
evaluates the information, and the patient, and makes final judgments about medication
management. The exact division of labor between NCC and prescriber is determined at the local
level. In the MHTS, the NCCs assumed responsibility for conducting clinical symptom and side
effect ratings, and providing patients with education about their medications and their illnesses.
Manuals provided by the investigators guided NCC medication decisions. The SMM experts trained
the NCCs on how to conduct clinical rating scales at orientation. Investigators periodically made site
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-5
visits to review and reinforce scale-rating skills of the NCC. Expert consultations were available by
phone and e-mail throughout the conduct of the study.
The SMM model provided to treatment group participants differs from current medication
management practices both in mental health clinics and in private practice settings in several
important respects. First, there are few nurses in mental health clinics or psychiatrists’ offices, and
their duties typically revolve around “medical” activities such as measuring vital signs and
administering injections. Second, monitoring medication-related outcomes and medication decisionmaking are typically under the purview of prescribers, without systematic input from other clinic or
office personnel. Third, responsibility for gathering and integrating current and historical
information on medication treatments, both psychiatric and non-psychiatric, falls to the prescriber.
Thus, in clinic and office settings, implementation of SMM requires systemic changes in personnel
roles and in recording of information.
Methods
This section presents the methods used to obtain psychiatric diagnoses and physical conditions in
the treatment group, and describes measures of SMM implementation and impact. Since the SMM
component was only available to treatment group participants, most of the relevant measures
collected by investigators were for that group only; thus, comparisons with the control group are not
possible.
Psychiatric Diagnoses
After beneficiary randomization into the treatment arm, MHTS investigators immediately began
gathering essential baseline data so that the SMM intervention could effectively begin. The
investigators needed to learn what the psychiatric diagnoses of the treatment group were and how
they related to SSA diagnostic categories. Trained clinicians obtained psychiatric diagnoses using the
Structured Clinical Interview for DSM-IV Axis I Disorders (SCID). These trained staff completed
the SCID as soon as practicable after randomization of a beneficiary to the treatment group. The
SCID interview consists of a series of questions designed to determine current and lifetime Axis I
diagnoses (See Chapter 2—Diagnostic Psychiatric Assessment) according to the 4th edition of the
Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) criteria. The structure of the SCID
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-6
interview is algorithmic, with initial screening questions followed by diagnosis-specific queries in
response to positive answers. The NCC and MHTS experts then used the SCID diagnoses as the
initial working diagnoses, which may have differed from the SSA diagnostic groups of
“schizophrenia” and “affective disorder” in SSA administrative files. The SCID interview also
identified other Axis I diagnoses, such as substance abuse or dependence and anxiety disorders, as
well as participants with no diagnosable Axis I disorder. SCID lifetime diagnoses are not mutually
exclusive from current SCID diagnoses. Therefore, it is possible to fall into both the
“schizophrenia” and “affective disorder” categories over the course of a participant’s lifetime. Cooccurring substance abuse is common in these disorders, and the SCID data allowed estimation of
the frequency of these and other co-occurring Axis I disorders in the treatment group.
Physical Health Conditions
An essential element in the study concerned the physical health conditions of participants and if or
how health-related issues affected their participation. Information about physical health conditions
came from a variety of sources. The eligibility screening process included an item that asked all
beneficiaries enrolling in the study about physical conditions that might interfere with work or study
participation. In addition, all participants indicated at baseline what medications they were taking.
The General Medical Exam (GME) was required of all participants randomized to the treatment
group. Since the goal of the GME was to exclude participants who were physically unable to work,
rather than to identify all medical conditions, there was considerable variability among providers as
to how extensively they documented conditions that were present but would not keep the
participant from working. Thus, in many cases the GME report did not list all current and past
medical diagnoses. The other source of information about medical conditions was the NCC, who
documented current DSM Axis III conditions (physical conditions) and all medications at intake.
The NCC also documented participant medications as part of the Comprehensive Transition Review
(CTR) for treatment group participants at the end of study participation.
MHTS investigators used medications as proxies for physical health conditions. Some medications
mapped easily onto putative physical conditions, while others were ambiguous. For example, the
team could easily map statins, used almost exclusively for the treatment of hyperlipidemias, while
propanolol in low doses has a variety of uses beyond treatment of hypertension. For purposes of
inferring physical conditions from medication treatments, the investigators adopted a conservative
approach to categorizing the medications taken by enrolled participants. The investigators mapped
medicines used for only one type of disorder (e.g., lung disease) to that disorder. In instances where
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-7
the medication had multiple uses, the team made no inference. Anti-epileptic medications, for
example, are widely used in the treatment of bipolar disorder, and inferring a seizure disorder from
their use would greatly overestimate the frequency of seizure disorders in this psychiatric population.
A listing of the medications and their corresponding mapped physical conditions is included in
Appendix 6A. If a report of a condition existed in the screener interview, GME, SMM intake form,
or through reports of medication use, the investigators determined a participant had a physical
condition. The sources for medication inference came from: (1) the Baseline interview, (2) the SMM
intake form, or (3) the end of study CTR form. Since these measures were frequently redundant,
researchers created an unduplicated count of participants with each condition. Most participants had
more than one physical condition.
As part of the SMM intake at baseline, NCCs recorded height and weight. The research team used
that data to calculate Body Mass Index (BMI) for each participant. Investigators grouped these
values into BMI ranges that define underweight, normal, overweight, and four levels of obesity.
Engagement in SMM
For investigators to examine the degree to which SMM implementation occurred and any
subsequent impact of SMM on employment outcomes, they needed to design strategies to monitor
quantity, quality and consistency of participant and prescriber engagement in SMM. Each NCC
completed a brief quality management (QM) assessment on each treatment group participant for
each quarter after SMM intake. One question asked if the participant had been actively engaged in
SMM during the previous quarter, based on the NCC’s contacts with the participant. Data from
these QM assessments were used to ascertain “ever” engaged (defined as having at least one
occasion when the participant was rated as engaged and “always” engaged (defined as the participant
being rated as engaged during every reporting period). MHTS investigators introduced the item on
SMM engagement six months after the study started; therefore, not every participant had the same
number of ratings. The data used in this analysis only included the 981 treatment group participants
who completed the two-year study period. This was done to avoid confounding of the definitions by
the 140 who were either administratively dropped from the study, formally withdrew, or died. When
the NCCs characterized treatment group participants as not engaged, they would choose among a
checklist of reasons for lack of engagement of the participant. The NCC could select as many
reasons as needed to describe the lack of engagement.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-8
In a separate questionnaire completed by NCCs in late 2009, NCCs were asked to rate on a fourpoint scale each of the prescribers, with whom they had worked, ranging from not at all engaged in
SMM to fully engaged in SMM. The questionnaire provided guidance with regard to criteria for each
rating. Additionally, the NCCs characterized each prescriber as providing services either “on-site” or
“off-site.” “On-site” did not necessarily mean the NCC and prescriber were co-located, but that they
operated within the same clinical system, used a common medical record, and could easily
communicate with each other in person, by telephone, or by a network e-mail system. Thus, for each
site, there were data about numbers of on-site and off-site prescribers and NCC perceptions of their
engagement in SMM.
At the beneficiary level, many participants had multiple prescribers over time, and some participants
moved between on- and off-site prescribers. For analytic purposes, the research team used the site
characterization of their last known prescriber in the study to determine if a prescriber was on- or
off-site, recognizing that this designation did not accurately capture the prescriber data history on
those with multiple prescribers over time.
NCC Role
At various points in the study, investigators requested that NCCs estimate time spent on different
functions, as well as to assess the relative value of the services they provided within SMM. An early
survey of NCC percent time and effort in various roles yielded widely divergent estimates.
Investigators speculated that non-recurring duties related to study start-up affected the wide
variation of responses. Toward the end of the study, investigators requested NCCs to estimate hours
per week spent on SMM and to evaluate their SMM role from various perspectives. Clinic
administrators provided survey data in the same period.
SMM Implementation
Each time a participant visited an NCC, it was the NCC’s job to complete an NCC clinical report.
Clinical need determined the frequency of these reports, except that investigators required NCCs to
complete an NCC clinical report for each participant at least quarterly. Hence, in the 2-year period
of study participation, the threshold minimum number of NCC clinical reports was set at eight
reports per beneficiary. Investigators considered that SMM implementation was inadequate if there
were fewer than eight recorded NCC clinical reports for a participant. Failure to meet this criterion
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-9
was due to various factors including the participant’s lack of engagement in SMM. Some sites were
without an NCC for periods ranging from one to five months. In addition, as mentioned in Chapter
2, two sites ceased active enrollment of participants within one year of study onset and discontinued
the staffing for both site research positions. Westat was able to provide NCC coverage for one of
these two sites allowing for the continued capturing of clinical research data. However,
the extended absence of an NCC affected the data on numbers of completed reports and clearly
affected SMM implementation during the period of absence. The quality of medication management
ratings were dependent on availability of NCC reports, of which there were none during any
extended absence of an NCC.
Prescriber reports in the SMS served as the comparable measure of prescriber engagement in SMM.
Investigators anticipated that prescribers would see each participant at least once every four months,
or six times in the course of two years. NCCs had no direct control over frequency of prescriber
visits and could only pursue strategies to encourage prescriber cooperation. NCCs typically received
completed prescriber reports after the participant’s prescriber visit. In some cases, the NCC
completed the prescriber report based on information gleaned from the prescriber and the medical
record.
Quality of Medication Management
Two study investigators (Drs. Bond and Miller) developed a rating scale to measure quality of
prescriber medication management of schizophrenia in a prior project (Taylor et al., 2009). For the
MHTS, in consultation with experts in treatment of depression and bipolar disorder, the scales were
adapted for assessment of quality of medication management of bipolar disorder and major
depressive disorder. Medical record reviews served as the basis for these quality assurance (QA)
ratings. The scales were somewhat lengthy and required the rater to search through the record
looking for evidence that the prescriber had documented intent and rationale for medication
decisions and had attended to patient symptoms and side effects. Research Assistants (RAs)
completed the scales on a 10 percent sample of participant records at each site each quarter. The
prime contractor randomly selected site records for review by way of an automated process of the
SMS. The RAs excluded previously selected records.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-10
Medication-Related Self Reports in Final Followup Interview
In the final Followup interview, interviewers asked participants if they were taking medication for
their psychiatric illness. If the participants responded “yes,” the interviewers asked followup
questions about frequency of taking the medication as prescribed, whether they had sufficient
information about the medication(s), and whether their attitude about taking the medication(s) was
positive, negative, or neutral. These questions were added to the final Followup interview because
part of the NCC’s role was to help treatment group participants understand the benefits of regular
medication adherence and to educate them about medications. Thus, these questions could
potentially help identify NCC effects on these behaviors and attitudes in the treatment group,
relative to the control group.
Results
This section reports descriptive analyses of SMM in the MHTS. The investigators did not emphasize
site differences in SMM here, but that information will be of value in multivariate analyses of site
differences in beneficiary participant and other outcomes.
Psychiatric Diagnoses for Participants in the Treatment Condition
SSA disability diagnostic categories of schizophrenia or an affective disorder served as selection
criteria for study participants. Current and past psychiatric diagnoses were determined through
SCID interviews. Appendix 6B presents a summary of the SCID diagnostic frequencies. The data
presented here are for all 1,025 persons assigned to the treatment condition and who completed a
SCID interview. This number includes participants who did not complete the two-year study
intervention. A comparison between the SCID diagnoses and the SSA diagnostic categories showed
87.92 percent concordance when including those with a dual diagnosis (Figure 6-4) and 80.33
percent concordance when excluding those with a dual diagnosis (Figure 6-5). Ambiguous case
classifications affected the degree of concordance. When the investigators rated the participants with
a SCID diagnosis of schizoaffective disorder or with diagnoses of both affective disorder and
schizophrenia as concordant (Figure 6-4), then concordance was somewhat higher. However, if
investigators limited its rating of schizoaffective disorder as only concordant with schizophrenia and
those with combined diagnoses were excluded from the analysis, then the level of concordance is
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-11
lower (Figure 6-5). Among the 39 participants without a schizophrenia spectrum or affective
disorder diagnosis based on the SCID, most had at least one psychiatric diagnosis. Only eight did
not have an Axis I SCID diagnosis.
Figure 6-4.
Concordance of SSA diagnostic group with SCID diagnosis (includes dual diagnosis)
100.00%
Legend:
SCID Diagnosis/SSA
Group
90.00%
80.00%
Percent of Total (N=993)
70.00%
Affective or Schizoaffective
/Affective
63.75%
Schizophrenia or
Schizoaffective/Schizophrenia
60.00%
Affective/Schizophrenia
50.00%
Schizophrenia/Affective
40.00%
30.00%
Other/Affective
24.17%
Other/Schizophrenia
20.00%
5.94%
10.00%
2.42%
0.00%
Concordant
(87.92%)
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
2.92%
Discordant
(12.08%)
6-12
0.81%
Figure 6-5.
Concordance of SSA diagnostic group with SCID diagnosis (excludes dual diagnosis)
100.00%
90.00%
80.00%
Legend:
SCID Diagnosis/SSA
Group
Percent of Total (N=946)
70.00%
60.00%
57.51%
Affective/Affective
Schizophrenia or
Schizoaffective/Schizophrenia
Schizophrenia or
Schizoaffective/Affective
Affective/Schizophrenia
50.00%
40.00%
30.00%
22.73%
Other/Affective
20.00%
9.62%
10.00%
0.00%
Concordant
(80.33%)
Other/Schizophrenia
6.24%
3.07%
0.85%
Discordant (19.67%)
It was common to find co-occurring Axis I psychiatric diagnoses in addition to affective disorders
and schizophrenia spectrum diagnoses (Table 6-1). The most common co-occurring condition was
substance abuse or dependence. Anxiety disorders were also common, especially among the group
with depression diagnoses. Both these observations are consistent with the psychiatric literature on
co-occurring disorders (Kessler, 2004; Grant et al., 2004). There were 12 (1.2%) diagnoses of a
General Medical Condition (GMC) resulting in a psychiatric condition (i.e., the psychiatric condition
was secondary or resulting from the GMC), and 10 of these were for participants who did not have a
diagnosable affective disorder or schizophrenia spectrum illness. Eight participants (0.8%) had no
SCID psychiatric diagnosis. Table 6-2 illustrates the frequency of the number of psychiatric
conditions in the MHTS treatment group. The SCID data refer to lifetime diagnoses. At any point in
time an illness might be in remission. However, each of these illnesses was recurrent; having had the
condition historically means that the person was considered at high risk for a return of active
symptoms.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-13
Table 6-1.
Substance abuse
disorders
Anxiety disorders
GMC
Other
None
Total (excluding
none)
1
2
3
4
5
Co-occurring psychiatric conditions for participants in the treatment condition
affective5
No
or
Schizophrenia4
diagnosis
(n=39)
Both affective5
and
Schizophrenia4
diagnosis
(n=55)
Total
(N=
1,025)
Depressive
disorders1
(n=3132)
Bipolar3
disorder
(n=300)
Schizophrenia4
(n=318)
128
79
1
2
140
138
38
0
2
137
147
23
1
0
157
15
12
10
3
8
26
5
0
2
26
454
157
12
9
468
210
178
171
40
33
632
Depressive Disorder or Depressive Disorder with Psychotic Features excluding those who also have Schizophrenia or Schizoaffective.
Fourteen (14) participants have both Depressive Disorder and Depressive Disorder with Psychotic Features. The duplication was
excluded from the total count for Depressive Disorders.
Bipolar Disorder excluding those who also have Schizophrenia or Schizoaffective.
Schizophrenia or Schizoaffective Disorders excluding those with an Affective Disorder.
Depressive Disorder or Depressive Disorder with Psychotic Features or Bipolar Disorder.
Table 6-2.
Frequency of co-occurring psychiatric conditions
Number of conditions
0
1
2
3
4
5
Total
Number of beneficiaries
8
447
481
82
6
1
1,025
Physical Conditions
Figure 6-6 shows the frequencies of the most common physical conditions for the 981 treatment
group participants who completed the two-year study period. Hypertension was the most common
condition, but there was a wide range of common medical conditions. Diabetes affected 207
beneficiaries (21.1%), a percent that was substantially above what one would expect in the general
population for this age group (22-56 years old).
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-14
Figure 6-6. Frequency of co-morbid physical health conditions for treatment group beneficiaries
The estimations of frequencies of physical conditions came from a variety of sources. Table 6-3
shows frequency of condition by data source. About half the conditions could be inferred from
current medication treatments (collected during the Baseline interview, SMM Intake, and
Comprehensive Treatment Review process), but this varied quite widely across conditions. Because
investigators restricted medication-based inference to medications whose use was largely limited to
one category of illness, the estimations likely underestimate actual incidences of physical conditions,
and of their medication treatment, to a modest degree. However, the current results support
investigators’ conclusions that a substantial number of beneficiaries had untreated physical
conditions known to have deleterious long-term health consequences (e.g., hypertension).
Figure 6-7 shows a classification of the number and percent of participants grouped by the number
of physical health conditions. The graph indicates that on average, there were between two and three
physical conditions for each participant in the treatment group and only one eighth of the group had
no identifiable physical condition.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-15
Table 6-3.
Physical health conditions by source (treatment beneficiaries) (N=981)
Condition
Anemia
Autoimmune Disorders
Blindness
Brain Damage
Cancer
Cardiovascular Diseases
Chronic Fatigue Syndrome
Chronic Lung Disorder
Chronic Pain Conditions
Diabetes
Gastrointestinal Disorders
HIV
Hearing Loss
Hyperlipidemia
Hypertension
Liver Disease
Migraines
Narcolepsy
Neuromuscular or Degenerative
Disorders
Neuropathy
Renal Disease
Seizure Disorder
Thyroid Disorders
Other
1
2
3
4
5
Eligibility
screener1
1
5
0
5
5
5
4
11
42
19
6
1
1
2
8
9
3
1
4
4
0
7
5
4
Baseline
interview2
13
8
0
0
6
20
0
80
147
93
139
19
0
137
222
0
12
0
41
15
19
21
84
0
SMM
Intake3
22
8
0
0
6
23
0
105
172
124
171
24
0
155
273
0
16
0
48
GME4
15
7
6
99
45
116
10
189
131
170
67
23
8
83
324
69
29
1
7
CTR5
20
7
0
0
8
28
0
106
151
134
175
21
0
203
285
0
9
0
52
Unduplicated
beneficiary
total
55
18
6
102
54
138
12
251
295
207
259
35
8
280
429
71
46
2
76
18
27
24
98
1
27
42
46
131
62
21
30
30
110
1
50
87
72
171
68
Beneficiary self-reported health conditions during the final eligibility screening process administered prior to enrollment.
Beneficiary self-reported medications in the Baseline interview.
Systematic Medication Management Intake conducted by the NCC. Based on beneficiary self-report and review of medical records.
Beneficiary health information based on the GME findings.
CTR of medications during Transition Planning.
Weight gain is a common side effect of psychiatric medications and obesity is a known health risk,
therefore, investigators tracked BMI. Figure 6-8 shows the findings on BMI by gender. Only 20
percent of participants were not overweight or obese. The majority of participants fell into the obese
range, with more females in the more severe obesity categories.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-16
Figure 6-7.
Frequencies of beneficiaries with one or more physical health conditions (N=981)
225
200
182
194
Frequency
175
158
150
125
122
125
100
84
75
56
50
28
25
16
16
0
0
1
2
3
4
5
6
7
8
Number of Conditons
Figure 6-8.
BMI distribution in treatment group by gender
0.0%
Underweight
<18.5
5.0%
10.0%
15.0%
20.0%
25.0%
35.0%
1.5%
0.7%
Normal 18.524.9
18.7%
19.1%
Overweight
25.0-29.9
20.8%
Obese 30.034.9
21.0%
Severely Obese
35.0-39.9
15.7%
14.9%
Morbidly Obese
40.0-50
Super Obese
>50
30.0%
9.4%
18.2%
4.0%
2.2%
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-17
29.4%
24.3%
FEMALE
N=523
MALE
N=456
≥9
Engagement in SMM
Participant utilization affects the effectiveness of any intervention. Appendix 6C presents a summary
of NCC quarterly SMM QM findings. Based on these assessments of treatment group participants,
most were engaged in SMM at some point during their study participation, and the majority were
engaged throughout (Table 6-4 excludes one site that ceased collecting all SMM QM research data
one year into the study). The range across sites in percent of beneficiaries that reported “always” or
“ever engaged” was especially pronounced for the “always engaged” category (25 – 88.6%). The
“always” category excluded QM assessments completed in the first 6 months of the study because
they did not include the engagement question.
Table 6-4.
NCC ratings of beneficiary engagement in SMM1
Ever
(any reporting period)
90.42%
(Site Range 75% – 100%)
Always2
(all reporting periods)
57.32%
(Site Range 25% – 88.6%)
1
Based on completed SMM QM Forms for the 981 treatment beneficiaries enrolled in the MHTS for 24 months. The number of QM
forms completed for each beneficiary varied, ranging from 0 to 8.
2
Excludes SMM QM Forms that did not contain the engagement question.
When completing the SMM QM templates for beneficiaries not engaged in SMM, NCCs were
required to record the reason for unengagement. The reasons most commonly given were “refusing
to meet with NCC to perform rating scales” (27%), “not on psychiatric medications” (10%), and
“cannot locate beneficiary” (7.5%). Appendix 6C includes a summary of reasons for nonengagement in SMM by site.
Toward the end of the study, NCCs completed a questionnaire rating prescriber engagement in
SMM, at a time when they could characterize almost all prescribers who saw study participants. The
differences in level of engagement between on-site and off-site prescribers strikingly favor on-site
prescribers (Table 6-5, p<0.001 by chi square). Multiple prescribers from other mental health
systems treated many of the MHTS treatment beneficiaries, accounting for the large number of offsite prescribers reflected in the report.
Table 6-5.
Relationship
On-site
Off-site
NCC ratings of prescriber engagement by prescriber location
Not at all
engaged (1)
freq
%
3
4.7
96
33.8
Minimally
engaged (2)
freq
%
8
12.5
107
37.7
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-18
Moderately
engaged (3)
freq
%
19
29.7
49
17.3
Fully
engaged (4)
freq
%
34
53.1
32
11.3
Total
64
284
About half of the beneficiaries were seen by on-site prescribers and half by off-site prescribers.
Table 6-6 reports the percent of beneficiaries seen by on- and off-site prescribers for each level of
prescriber engagement. There were more missing engagement ratings of off-site than of on-site
prescribers. Since off-site prescribers were more likely to be minimally or not at all engaged, the data
in Table 6-5 likely underestimates the proportion of beneficiaries seen by prescribers in these two
categories.
Table 6-6.
Relationship
On-site
Off-site
Total
1
2
Percent of beneficiaries seen by prescribers (at final study visit by beneficiary)
grouped by NCC ratings of prescriber engagement and by prescriber location
Not at all
engaged
freq
%1
5
0.7
82
11.0
87
11.6
Minimally
engaged
freq
%1
37
4.9
129
17.2
166
22.2
Moderately
engaged
freq
%1
69
9.2
8.2
61
130
17.4
Fully
engaged
freq
%1
334
44.7
31
4.1
365
48.8
Total2
445
303
748
The denominator for % Total is the Total Non-Missing Beneficiary Ratings N=748.
There were a total of 226 beneficiaries with a missing prescriber rating; 47 on-site and 179 off-site prescribers.
NCC Roles
In the NCC administered survey conducted in 2009, one question was about number of hours per
week spent on SMM. More than 83 percent of NCCs estimated 10-30 hours per week, with slightly
more than half this group reporting 10-20 hours per week spent on SMM.
In the same survey, NCCs identified which of their activities they thought were of greatest benefit to
patients. The most frequent choice by far was “providing continuity of care” (ranked first or second
most important by 87% of respondents), with “tracking medications,” and “performing psychiatric
scales” also being endorsed moderately frequently. Continuity of care relates to the NCCs role of
integrating care across providers, including the treatment team, prescribers, and physical health care
providers.
When asked if NCCs can effectively perform their jobs with on-site prescribers, 76.7 percent
reported “most of the time” or “all of the time.” For off-site prescribers, only 16.7 percent of NCCs
indicated that they could effectively work with this group “most of the time” or “all of the time.”
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-19
Twenty-one administrators responded to survey questions about the role of the NCC. The large
majority endorsed the importance of the function and rated it as having greater value than traditional
nurse functions. However, more than 75 percent indicated very little likelihood that there would be
funding available for such a position after the study ended. Appendices 6D and 6E present a
summary of the findings from the NCC Survey and the NCC Survey Administrator Version.
SMM Implementation
The measures of implementation assessed frequency of required documentation of participant visits
to the NCCs and prescribers. Table 6-7 shows the frequency with which NCCs and prescribers met
minimal thresholds for documentation of participant visits. It is clear that NCCs were much more
likely to achieve these threshold levels of visit frequencies for participants. Appendix 6F presents a
summary of SMM implementation findings by site.
Table 6-7.
SMM implementation (% of participants1)
Across all sites
Site range
NCC form number above threshold2
80.8
8.3 – 100.0
Prescriber form number above threshold3
37.2
0.0 – 93.5
1
2
3
Based on the 981 participants enrolled in the intervention for two years.
The minimum standard for NCC Clinical Reports was eight completed forms during the 2-year study period (i.e., one form at least once
every three months).
The minimum standard for Prescriber Reports was six completed forms during the 2-year study period (i.e., one form at least once
every four months).
Quality of Medication Management
The record review to evaluate quality of medication management (SMM QA) contained 23 items,
each rated on a 5-point Likert-type scale. Not all items are applicable to all participants (e.g.,
treatment of refractory symptoms).
The item mean scores ranged from 2.0 to 4.9. Above or equal to 4 is good, 3-3.99 may be
problematic, depending on the measure, and 1-2.99 indicates a definite need for improvement.
Appendix 6G shows a summary of the SMM QA findings by site.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-20
The grand mean of the item score means was 3.76 for the 265 treatment group participants who had
ratings of their charts. Items with poor scores, on average, were documentation of side effects,
description of side effects of current medications, evidence that outcome measures were used to
guide treatment, and evidence of review of side effect treatments. These are all items that depend on
prescriber documentation in the medical record.
Mean scores for the study sites on the scale as a whole ranged from 2.9 to 4.5. Since each site had its
own rater, there could have been some degree of systematic scoring differences across sites.
However, investigators did spot checks of ratings of quality of medication management during site
visits, and provided feedback to raters to reduce variability across sites.
Medication-Related Self Reports in Final Followup Interview
Results from the medication-related items included in the final Followup interview (Table 6-8)
revealed that both groups had very high rates (>93%) of reporting that they took their medication as
prescribed most of the time and that they had sufficient information about their medication(s).
There were no differences between the treatment and control groups on these two questions.
With regard to attitude toward the use of psychotropic medications to control symptoms, nearly
two-thirds of participants in the treatment and control groups reported positive attitudes. There
were no group differences.
The groups did differ significantly in the proportion of respondents who reported currently taking
psychotropic medications, with the treatment group reporting a higher current medication use
(4.3%). This difference was statistically significant (p-value = 0.01). Since this was at the end of the
study, this difference might be attributable to participation in SMM having prompted more
participants to take medication.
Discussion
The results presented in this chapter point to a number of conclusions of importance to SSA. The
discussion follows the same order as the presentation of the results.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-21
Table 6-8.
Summary of medication-related self-reports in final Followup interview
Interview Question
Are you currently taking any
prescription medications for an
emotional or mental problem, or a
problem with your nerves?
Yes
No
Missing
Treatment
(N=902)
freq
%
Control
(N=991)
freq
%
Total
(N=1,893)
freq
%
778
122
2
86.3
13.5
0.2
813
177
1
82.0
17.9
0.1
1,591
299
3
84.0
15.8
0.2
739
28
11
0
95.0
3.6
1.4
0.0
773
31
7
2
95.1
3.8
0.9
0.2
1,512
59
18
2
95.0
3.7
1.1
0.1
724
47
7
93.1
6.0
0.9
776
34
3
95.4
4.2
0.4
1,500
81
10
94.3
5.1
0.6
503
98
171
6
64.7
12.6
22.0
0.8
545
109
147
12
67.0
13.4
18.1
1.5
1,048
207
318
18
65.9
13.0
20.0
1.1
How often do you take your
psychiatric medications as
prescribed by the doctor or as
directed on the label?
Most of the time
Some of the time
Less than half the time
Missing
Do you have all the information you
need about your psychiatric
medications?
Yes
No
Missing
In general, how do you feel about
taking psychiatric medications?
Positive
Negative
Neither
Missing
Concordance between SSA diagnostic category and SCID diagnosis was at least 80 percent. While
this observation helps validate the diagnostic basis for disability, there was only a weak correlation
between diagnosis and functional impairment. Use of the SCID in psychiatric disability evaluations
would undoubtedly improve diagnostic precision and might improve treatment selection. However,
the contribution of improved diagnostic capabilities to decisions about level of disability is arguable.
Many studies have shown that co-occurring psychiatric conditions, especially substance abuse,
worsen the course of illness of affective and schizophrenia spectrum disorders. Concurrent
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-22
treatment of both disorders is considered to be far preferable to sequential treatment (Drake &
Bond, in press). With 44 percent of the treatment group having a lifetime diagnosis of substance
abuse or dependence, the value of co-location of mental health and substance abuse treatment
programs was evident. In the MHTS, treatment of substance abuse problems was left to the sites.
Results did not indicate significant improvements in this area (Chapter 4), but the frequency of
treatment for substance abuse was only 13 percent (Chapter 5). Future work should examine the
effects on employment with greater focus on substance abuse, using evidence-based treatment
strategies.
The added burden of chronic physical health conditions affects functional abilities directly through
the physical health problem, as well as via side effects of treatments for these problems. More than
87 percent of the treatment group had at least 1 physical health condition, according to estimates,
and 69 percent had 2 or more. Moreover, more than half the group had a BMI in the obese range.
All of these observations are in accord with recent reports that persons with serious mental illnesses
have life spans that average up to 25 years shorter than the rest of the population and that this
premature mortality is mostly attributable to physical health conditions (Parks et al., 2006).
The NCCs reported that for a number of participants impairment from physical health conditions
overshadowed the mental health condition that had originally been the basis for their disability.
Efforts to increase employability of the population with mental health disabilities should attend to
medical as well as mental health issues. The MHTS did not directly measure physical disability (the
SF-12 physical component assesses patient self-perception of physical health) and did not
systematically record level of control and severity of most physical conditions during the course of
the study. Hence, there are no data on the impact of SMM on direct measures and treatment of
physical conditions.
Field reports from the NCCs and from the Quality Management Project Directors suggest that the
role of the NCCs in identifying physical conditions, in recording the medications used to treat them,
and in contacting physical health care providers was highly valued as an important part of the
treatment program. In principle, the prescribers of psychotropic medications for study participants
would be fully aware of their physical conditions and the treatments for them. The reality of our
fragmented care system, however, is that care for mental and physical health care conditions is
typically not integrated, often resulting in the neglect of physical health care for the mentally ill. In
the MHTS, the NCC’s role was essential to the integration of mental and physical health care, and in
helping participants get their physical health care needs met.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-23
Engagement of prescribers and of participants in SMM was quite variable, as was degree of
implementation. The data are clear that working with off-site prescribers is extremely challenging in
implementing SMM. For purposes of conducting the MHTS, the decision to allow participants to
remain with their outside prescribers was perfectly reasonable. However, the goal of the MHTS was
to deliver an integrated package of interventions to participants; having an off-site prescriber
presented great difficulties in integrating the SMM component with other treatments.
Chart reviews showed that quality of medication management varied considerably across sites and
items. The poor documentation related to side effects is very consistent with literature reports
(Cradock et al., 2001). Again, the off-site location of many prescribers affected this measure, since
these prescribers were much less likely to use the documentation forms recommended in the MHTS.
Results indicate areas of poor performance that could be addressed directly by the NCC (e.g.,
assessment and documentation of side effects), or indirectly through more aggressive prompting of
prescribers (attending more to outcome measure ratings in making medication decisions).
The final Followup interview did not find group differences in reported adherence to, knowledge
about, and attitudes toward psychotropic medications among those taking them. With both groups
reporting more than 90 percent good adherence and good knowledge, there may have been a ceiling
effect on these items. It is worth noting that in similar patient populations, actual “good” adherence
rates are typically 60-70 percent, and that patients usually overestimate their own adherence (Velligan
et al., 2007). Direct measures of adherence and of knowledge would have been helpful, but were
beyond the scope of the interviews.
Positive attitudes toward medications also did not differ across groups. The interviewer only asked
this question as a followup of those who indicated they were taking psychotropic medication(s). It
might be more revealing in future surveys to evaluate these attitudes in those not taking medications,
of which there were more in the control group.
Finally, there is the question of feasibility of SMM implementation in existing mental health systems.
Nurses are relatively expensive staff. There may well be cost savings from the NCC contributing to
greater prescriber efficiency, but it is not clear that the position can pay for itself under current
reimbursement schemes. While SSA does not pay directly for mental health services, implementation
of the service package used in MHTS for persons with schizophrenia or an affective disorder will
require other governmental agencies to develop policies and procedures that promote and pay for
these services.
Chapter 6: Implementation of the Nurse Care
Coordinator Role and Systematic Medication
Management
6-24
Chapter 7
Health Care and Supported Employment Financing
A primary goal of the Mental Health Treatment Study (MHTS) was to ensure treatment participants
had access to needed health care and supported employment (SE) services and supports that
facilitated their return to competitive employment—without fear of incurring burdensome out-ofpocket expenditures for health insurance premiums, co-pays, work-related expenses, and other
services. Toward this goal, Westat set out to identify and develop individualized benefit packages
that optimized the coverage of health care, SE, and other service needs for each participant while
minimizing participants’ out-of-pocket expenditures. One primary challenge for the study was
ensuring availability of needed evidence-based services; a second was providing the wherewithal
needed to access those services. Chapters 5 and 6 addressed the level of success in making the
services available to participants. This chapter describes the strategy for gaining access to those
services.
The team first identified the categories of health care, SE, and other services that participants in the
treatment group would require when participating in the intervention. The team then sought
additional information to determine the expected spending for these services and related health care
coverage and identified the relevant financing mechanisms.
This chapter presents a summary, in broad scope, of the financing principles and key decisions that
cumulatively formed the financing framework for providing treatment intervention participants
complete access to the services they needed. Health care and SE services that constitute the benefit
package referred to as the Health Benefits Plan (HBP) by the Social Security Administration (SSA).
Health Care and SE Financing Principles
Westat developed the following eight principles that guided the development of health care and SE
financing for individual participants and related payment decisions. SSA study design requirements,
recommendations from the Technical Panel, and information gathered by Westat served as the
foundation for these principles.
1.
Provide all participants a comprehensive health care coverage package. The
Introduction and a section in Chapter 2 describe the MHTS intervention components.
Chapter 7: Health Care and Supported
Employment Financing
7-1
Study Design and Methodology necessitated that participants had access to traditional
health care services (i.e., medications, outpatient and hospital or inpatient services for
both behavioral health and general medical conditions). In addition, participants needed
access to SE and related support services. Consequently, all participants in the treatment
group needed comprehensive health benefit packages that, at a minimum, provided
coverage for medications, outpatient care, inpatient care, and SE.
2.
Ensure that participants receive needed care. Not all insurance plans cover
behavioral health treatments recommended in a participant’s MHTS treatment plan. For
example, Medicare did not cover certain types of rehabilitative services. When the
treatment plan called for a particular treatment—whether a service, medication, or
otherwise—lack of insurance coverage for the treatment was not to preclude the study
covering the needed treatment.
3.
Ensure that participants receive evidence-based treatments. All participants were
to receive evidence-based SE services via the Individual Placement and Support (IPS)
approach. Likewise, participants were only to receive evidence-based behavioral health
treatments, including systematic medication management (SMM).
4.
Cover SE in full. Because SE represented an essential MHTS intervention for all
treatment participants and health insurance did not necessarily cover this service, the
study covered the cost of SE in full. SE was not a covered benefit under Medicare and
was almost never (if ever) covered by private health plans. Medicaid sometimes covered
SE services in full or in part, depending on the state.
5.
Cover all other approved MHTS behavioral health care expenditures in full.
Participants should never become liable for any part of the expenditures of a MHTSapproved treatment for a behavioral health condition. (MHTS-approved treatments
were services indicated in the participant’s MHTS treatment plan or medications
prescribed by the participant’s prescriber.)
6.
Utilize scarce resources wisely. Westat was mindful that both within the MHTS and
in the broader health care market, health care resources were scarce. Furthermore, SSA
limited MHTS health care and coverage funds to an average of $10,000 per participant
for each year they were in the study (24 months). Thus, it was imperative that the study
employ the limited resources as efficiently as possible. To this end, the study simplified
the process of coordination of benefits, payment of claims, and reimbursement of outof-pocket expenditures.
7.
Minimize participant up-front, out-of-pocket expenditures. The reimbursement
process developed by the investigators minimized participants’ up-front out-of-pocket
spending. In addition, the study later reimbursed these expenditures. Westat recognized
that guaranteeing reimbursement for approved out-of-pocket expenditures might not
have overcome the financial barriers to treatment. Many MHTS participants had limited
financial resources and thus were unable to produce the credit or cash on-hand
necessary to purchase needed medications or services, even when they knew that the
study would reimburse them. Such participants simply could not pay the required out-
Chapter 7: Health Care and Supported
Employment Financing
7-2
of-pocket expenditures up front. Westat staff worked with the site staff to develop
service payment solutions for such situations.
8.
Collect research data on encounters and expenditures. Encounter and service
expenditure data were essential in understanding the overall spending during the
implementation of the MHTS intervention components across sites. Westat
implemented payment mechanisms to support the timely collection of accurate
encounter and service expenditure data in order to inform policy on this population and
to provide the services participants needed to recover, achieve independence, and work
competitively.
Westat drew upon various sources to estimate the expenditures associated with the needed services
and supports for MHTS participants and the price of a standard health care benefits package. The
MHTS Health Benefits Plan team obtained information that included actuarial estimates of
insurance premiums, the availability of comprehensive health insurance policies for the uninsured,
historical spending on SE, and Medicare Part D policy (for prescription drug coverage). Members of
the team of investigators conducted focus groups with MHTS site directors to learn how the sites
customarily finance the SE and behavioral health services they deliver.
Health Care Coverage and Financing Decisions
This section chronicles the fact finding process and summarizes the history, the decisions made,
methodology used, and the payment mechanism used to ensure appropriate health insurance
coverage and premium reimbursements for treatment participants during the course of their
participation in the treatment intervention.
Participants’ Expected Insurance Coverage at Time of MHTS Enrollment
While the prevalence of participants with private plans was unclear, based on findings from previous
studies of adult populations with disabling mental disorders, the team anticipated that:

Approximately one-third of MHTS participants would be ―dual-eligible‖ (i.e., have both
Medicaid and Medicare coverage), having qualified first on Social Security Disability
Insurance (SSDI) but later having qualified for Supplemental Security Income (SSI);

Nearly two-thirds would have Medicare only;
Chapter 7: Health Care and Supported
Employment Financing
7-3

Approximately one percent would have neither coverage (i.e., be in the 24-month
Medicare waiting period); and

One percent would be in the waiting period for Medicare, but have Medicaid.
Insurance premium expenditures. The Hay Group (an MHTS subcontractor) provided Westat an
actuarial estimate of the premium cost for commercial health insurance that provided individual
coverage equivalent to the benefits provided under Medicare Parts A, B, and D; the amount was
$3,780 per year or $315 per month. Westat used this estimate of the cost of comprehensive coverage
as a guide in determining the MHTS premium reimbursement amounts for participants with various
insurance coverage arrangements.
Availability of health insurance policies for the uninsured. If the study were to enroll
participants possibly with no health care coverage (e.g., those in the 24-month Medicare waiting
period), Westat needed to purchase coverage equivalent to Medicare Parts A, B, and D for these
participants as soon as possible after enrollment. Consequently, a Westat team of experts in health
economics and mental health service policy researched the availability of comprehensive health
insurance policies for participants in each of the states in which study sites were located. In all states
except Maryland, the available insurance plans contained stringent pre-existing condition clauses.
These policies did not cover treatment for any condition that the participant had at the time he or
she enrolled in the insurance plan until a waiting period ranging from 3 months to 24 months had
passed. Thus, the plans precluded coverage for participants’ (well-documented) mental disorders, as
well as for any general medical conditions they had when they enrolled in the study.
The two state-operated sites (both in Connecticut) proved an exception. Since the state itself
administered these two sites, they agreed to provide behavioral health care, both outpatient
(including SMM) and inpatient, to uninsured MHTS participants at the state-run facilities at no cost.
As with other sites, the study compensated these two sites for SE and related services.
Medicare Part D policy. Members of the team of investigators, Westat health economists, and
University of Texas pharmacy experts studied the policies and regulations of the Medicare Part D, as
well as costs and cost sharing for a sample of Part D plans. Study results found that Part D plan
premiums ranged from $15 to more than $100 monthly, depending on many factors, including the
plan’s cost-sharing requirements, the specific medications on the plan’s formulary, whether coverage
was provided in the plan’s so-called ―donut hole,‖ and other factors. The ―donut hole‖ refers to a
gap in Part D coverage that occurred when the participant reached the plan’s initial annual coverage
limit, generally $2,400 in 2007. Such a participant has ―fallen into the donut hole‖ or coverage gap
Chapter 7: Health Care and Supported
Employment Financing
7-4
and becomes responsible for the total spending on all medications until total out-of-pocket expenses
on formulary medications for the year, including the deductible and initial co-insurance, reaches
$3,850. When total out-of-pocket costs reached $3,850, Part D coverage resumed.1 Some Part D
plans offered coverage ―in the gap,‖ while others did not. It became apparent that participants with
many or high-cost medications—whether for behavioral health or general medical conditions—
would represent a sizable expenditure to the MHTS when they hit the donut hole.
Based on the foregoing review and substantial discussion with experts, the following decisions
became the guiding blueprint for providing participants access to the intervention treatments and
services.
Decision 1: Provide comprehensive coverage equivalent to Medicare Parts A, B, and D. To
maximize access to effective treatments, all treatment participants received insurance coverage
equivalent to Medicare Parts A, B, and D upon enrollment in the MHTS. This level of coverage is
―comprehensive coverage.‖ Medicare Part A provides inpatient coverage, while Part B covers office
visits, lab work, and other office-based care. Part D provides prescription drug coverage. Thus,
Medicare Parts A, B, and D represent a comprehensive health insurance package and cover the
major services needed by MHTS participants—with the exception of SE, which the study covered in
full.
The MHTS provided participants who had not yet completed the Medicare 24-month waiting period
with comprehensive coverage (i.e., coverage equivalent to Medicare Parts A, B, and D). As discussed
earlier, because of the issues related to obtaining comprehensive coverage for the uninsured
participant, the MHTS sample initially included participants in the 24-month Medicare waiting
period only from the state of Maryland and the two state-operated sites.2 In the case of individuals
who had Medicaid, private, or other insurance at the time of enrollment, Westat tailored their
coverage to ensure its equivalence to Medicare Parts A, B, and D. For participants who chose to
remain on their existing private insurance policies, Westat met the MHTS comprehensive coverage
standard without compromising the individual’s preference, whenever possible. In short, Westat
brought participants’ insurance coverage up to the study standard regardless of their coverage at the
time of study enrollment for the duration of the 24-month intervention period.
1
Equivalently, Part D coverage resumed when the combined annual amount of the participant’s total out-of-pocket costs and Medicare payments
exceeded $5,451.
2
Enrollment for persons in the 24-month Medicare waiting period at other sites were postponed for 6 months until completion of a review of
enrollment data and further investigation into health care coverage options in the remaining study sites.
Chapter 7: Health Care and Supported
Employment Financing
7-5
For participants that entered the study with Medicare Part A (which requires no premium) but
without Part B or D (which require monthly premiums), Westat obtained Part B and Part D
coverage for them at the earliest opportunity (as needed), which often required waiting until the
respective open enrollment period. Part B open enrollment is from January 1 through March 31 of
each year, with coverage taking effect July 1 of that year; Part D open enrollment is from November
15 through December 31 of each year, with coverage taking effect the first day of the new year.
Regardless of any waiting periods, the participant did not incur a cost for MHTS-approved
behavioral health care treatments or medications. The study paid the full expenditures related to
such care when no other coverage existed.
Table 7-1 shows the insurance coverage that Westat obtained for each participant, depending on the
type of coverage the participant initially brought to the study, as well as the expected costs to the
study and to the participant for premiums, behavioral health care, and general medical care. As this
table shows, participants using MHTS-approved providers for MHTS-approved services incurred no
costs for their behavioral health care.
Decision 2: Reimburse premium payments for coverage equivalent to Medicare Parts A, B,
and D. Westat paid the insurance premiums for Medicare Parts B and D (Part A has no premium)
or equivalent coverage (i.e., comprehensive coverage). Whenever possible, Westat paid the insurance
premiums directly to the insurer to minimize out-of-pocket expenditures to participants. When this
was not feasible, Westat reimbursed the participants for their insurance premiums in advance of
premium due dates. For example, when participants had Medicare Part B coverage, the insurer is the
Centers for Medicare & Medicaid Services (CMS). CMS deducts premiums directly from Social
Security benefit checks. To avoid a lag between this premium deduction and participants’ receipt of
premium reimbursement checks, Westat paid premium reimbursements to these participants prior to
receipt of their Social Security benefits checks. Further detail to the approach to premium payments
is presented in greater detail in the next section covering Implementation of the HBP.
Chapter 7: Health Care and Supported
Employment Financing
7-6
Table 7-1.
Participant
group1
General
medical care
Behavioral
health
care2
medium
$0
None
Medicare
premiums
medium
low
$0
None
$0
very low
very low
$0
Medicaid only
None
$0
very low
very low
$0
51
No insurance
Coverage
equivalent to
Medicare
Parts A, B,
and D4
Higher than
Medicareequivalent
premiums
high
medium
$0
61
Private
insurance
None
Higher than
Medicareequivalent
premiums
high
low
$0
41
4
Westat
behavioral
health care
reimbursement2
high
3
3
Participant-paid
out-of-pocket costs
Medicare
premiums
2
2
Participant
health care
coverage at
study entry
Medicare
Parts A, B, and
D only
Medicare
Parts A, B, and
D
Medigap
Medicare
Parts A, B, and
D
Medicaid
Basic health
care coverage
Westat
obtained
following
Westat
basic
premium
coverage
cost3
None
1
1
Insurance plans and estimated costs to Westat and participants
Participants in groups 4, 5, and 6 are in the 24-month waiting period for Medicare.
The MHTS paid for all behavioral health care out-of-pocket costs incurred by participants using MHTS-approved providers for MHTSapproved services.
The MHTS paid premiums for basic health care coverage, Medicare Parts A, B, and D or the equivalent coverage.
If a participant with Medicare did not have either Part B or Part D coverage at study entry, the MHTS obtained Part B and the most
appropriate Part D policy for the participant and paid the premiums.
Decision 3: Delay enrollment of uninsured participants until costs are analyzed. After
analyzing the health insurance landscape for uninsured participants in states with MHTS study sites,
Westat concluded that the available health insurance policies would afford MHTS participants
insufficient coverage due to the pre-existing condition exclusions.3 Moreover, Westat had particular
concern about the lack of control over cost containment for long-term behavioral health care
3
Although the health plan may accept enrollment, SSDI beneficiaries who certainly have pre-existing conditions may not have coverage for any care or
services related to the pre-existing condition. Depending on the policy and the state’s insurance regulations, this exclusion period can range from 3 to
24 months.
Chapter 7: Health Care and Supported
Employment Financing
7-7
inpatient costs. A relatively small number of uninsured participants incurring lengthy inpatient stays
could potentially drain study intervention resources.
As noted earlier, preliminary research indicated that only in one state (Maryland) could the study
purchase comprehensive health care coverage—without pre-existing condition exclusions—for
uninsured participants. In addition, Connecticut’s two state-run MHTS study sites agreed to provide
behavioral health care to uninsured participants at no cost. Prior to enrollment, uninsured
beneficiaries who would be served at the state-operated study sites were informed that behavioral
health services, including inpatient care, could only be obtained through the MHTS study site. If
participants chose to receive behavioral health care outside of these facilities, they incurred the costs.
Finally, for uninsured participants from these state-operated sites, the study purchased insurance
coverage from the state’s high-risk pool to provide coverage for their general medical care.
For other than the exceptional accommodations mentioned above, Westat and SSA decided to delay
enrollment of participants who are in the 24-month waiting period for Medicare for 6 months until
completion of a review of enrollment data and further investigation into health care coverage
options for uninsured participants in the remaining study sites. After reviewing the collected data,
Westat and SSA decided to enroll all beneficiaries across all the study sites, regardless of the length
of time they had been receiving SSDI benefits. This addition became effective August 1, 2007, with
the provision of a spending cap on study expenditures for uninsured treatment group participants
from the remaining study sites.
Implementation of the HBP
Westat created a dedicated full-time position, the Westat Insurance Planner (WIP), to assess and
verify required insurance premiums, investigate additional insurance needs, facilitate obtaining any
additional insurance needed to attain a comprehensive level of coverage, and coordinate the timely
payment of insurance premiums. Participants’ insurance issues typically required immediate
assessment and response to ensure adequate coverage at all times during their study enrollment
period. The WIP responded to and resolved participant insurance premium and other coverage
matters in a manner that would not have been possible were the premium payment function housed
outside of Westat.
Chapter 7: Health Care and Supported
Employment Financing
7-8
Assessment and Verification of Existing Coverage
The Research Assistant (RA) at each study site collected detailed insurance coverage and premium
information on each newly enrolled treatment participant during the post-randomization meeting
(discussed in the Chapter 2). The Health Insurance Questionnaire (completed by the RA during the postrandomization meeting; Appendix 7A) showed the insurance information collected. The RA used
the secure MHTS fax line to send the WIP a copy of the completed Insurance Questionnaire and
copies of the participant’s insurance card(s). The WIP reviewed this information, collected additional
information as necessary, and established the extent of the participant’s current coverage; the WIP
then determined whether the participant needed additional coverage. During this process, the main
goal was to make sure that each participant had comprehensive health care coverage (i.e., Medicare
Parts A, B, and D or equivalent coverage from another source). For example, beneficiaries needed
Medicare Part B coverage if they did not have Part B or coverage from another source that paid for
physician services. Similarly, a beneficiary needed Part D coverage if he or she did not have
Medicare Part D or any other coverage for prescription medications.
To ensure the appropriate use of study funds for insurance premium reimbursement, the WIP
verified each participant’s insurance coverage to determine the level of reimbursement. In the course
of the study and especially during the enrollment phase, treatment participants often did not have
their insurance cards or Proof of Income Letter from SSA to verify their monthly premium
payments for Medicare Parts B and D. The Proof of Income Letter lists SSA benefits and any
deductions such as Medicare Part B or D premiums (if applicable). The letter would indicate no
premium deductions for participants who were dually eligible (Medicare and Medicaid) since the
premiums were paid by Medicaid. Westat, with significant support from the site RAs, relied on the
participants to supply the needed documentation. Trained site RAs helped participants request
necessary information from the SSA and Medicare websites. The WIP routinely visited the
Medicare.gov website to verify participants’ enrollment in Medicare Part D. Though sometimes
dated, the Medicare.gov website provided valuable information on participants’ Medicare Part D or
HMO coverage. In addition, the site provided premium cost information coverage limitations. Third
party queries (TPQY) were another verification tool used when allowed access by the state in which
a treatment participant resided. The TPQY provided a listing of all insurance premium deductions
along with effective dates. To verify private insurance premiums costs, a copy of the policyholder’s
paystub was requested (most often the participant was not the primary on the policy). In the event
that an employer covered the full cost of an insurance premium, the primary insured obtained a
letter from his or her human resources department. Once verified, the WIP entered this information
Chapter 7: Health Care and Supported
Employment Financing
7-9
in the insurance section of the Study Management System (SMS). The investigators were not
provided access to the CMS system for insurance verification. The process therefore was longer than
originally anticipated, taking on average four to six weeks per beneficiary.
Final notices were sent to participants via certified mail after six months where attempts to gather
the verification documentation failed. The letter outlined the failed attempts to collect this
information and listed the information needed. Some participants immediately responded by
providing the missing information. In these instances, the WIP updated the SMS and set up a
payment schedule. The WIP closed the insurance verification file and made a notation in the SMS
for participants who did not respond to the certified letter. The investigative team concluded that
most of these cases were individuals who were on Medicaid and did not pay for insurance
premiums. The study did reimburse any participant who subsequently produced proof of premium
payments.
Meeting insurance needs. Most participants entering the study had coverage through Medicare
only, Medicare and private insurance, or some other form of insurance (e.g., VA benefits or other
government benefits). However, some participants were within the 24-month waiting period to be
eligible for Medicare coverage and were thus uninsured. Other participants had Medicare Part A
only. Still others had Part A and D coverage or Part A and B only. The following address the steps
taken to enroll participants in Medicare Part B or D or another form of government insurance.
Lack of Medicare Part B coverage. Open enrollment for Medicare Part B occurs annually from
January 1 through March 31, with coverage becoming effective on July 1. In January of each year,
the SMS generated a report that listed the participant ID numbers for those who did not have
Medicare Part B coverage. The site RA met with each participant without Medicare coverage to
discuss insurance enrollment. The WIP provided each RA with the proper forms and instructions
on how to enroll any participant interested in enrollment. The RA documented the participants’
decisions about opting to enroll or not and communicated this information to Westat. Some
participants did not wish to enroll because they received coverage under a spouse or parent’s
insurance. Others simply refused. Participants who chose to enroll received confirmation, premium
cost information, and an updated Medicare card from CMS. Westat staff monitored the progress
and status for all affected participants throughout the open enrollment period and documented any
changes in the SMS system. Once SSA began to deduct premium costs from a participant’s SSDI
monthly payment, Westat generated a payment schedule in the SMS, to reimburse the participant for
the cost of that premium. The process is explained in detail below.
Chapter 7: Health Care and Supported
Employment Financing
7-10
Lack of Medicare Part D coverage. Open enrollment for Medicare Part D is November 15
through December 31 annually. In November of each year, the SMS generated a report listing the
participant ID numbers for those without Medicare Part D coverage. As with Part B enrollment
efforts, Westat provided each site RA with a list of the IDs for their participants without a
prescription plan. The site RA and the WIP assisted participants opting for enrollment or those
wanting to switch from their current plan utilizing the Medicare.gov website. The criteria for
choosing the Optimal Medicare Part D plans were as follows:
1.
Participant had neither medication history information nor Part D plan. When
no medication information was available at the time of Part D enrollment, the strategy
was to use the list of most commonly prescribed (brand-name) medications for the
following:
–
Schizophrenia: Seroquel, Zyprexa, Risperdal;
–
Bipolar: Depakote ER, Trileptal, Geodon; and
–
Depression: Zoloft, Lexapro, Effexor XR.
By using this method, there was consideration for drug expense and most commonly
prescribed behavioral health drugs (i.e., second generation antipsychotics, selective
serotonin reuptake inhibitors, and mood stabilizers) for the study population. Given this
medication list, the Medicare Drug Plan Finder (MDPF) produced a list of plans with
expected annual spending. Considerations for an optimal plan choice included:
2.
–
Expected total annual spending,
–
Coverage in gap,
–
Monthly premium and annual deductible,
–
Network of pharmacy, and
–
Rating (two or more stars in all plan quality dimensions).
Participant had medication history information but no current Part D plan.
Strategy was based on:
–
Current medication information; or
–
Predicted future medications (strategy 1) to find an optimal plan for the
participant with respect to the five items listed above.
Chapter 7: Health Care and Supported
Employment Financing
7-11
3.
Participant had a Part D plan but requested a switch of plans. Strategy was based
on:
4.
–
Current medication information; along with
–
Predicted future medications (strategy 1), to find an optimal plan for the
participant. The WIP compared the annual total spending of the optimal plan
with the participant’s current plan. To recommend switching plans the optimal
plan expected spending had to be at least 10 percent lower than the current plan.
Participant had private insurance plan that covered medication. In this case, the
study did not enroll the participant in a Medicare Part D plan.
To avoid a participant incurring unwanted Part D premium plan payments during the calendar year
in which their study participation ended, the MHTS made the decision to continue Medicare Part D
reimbursement for the entire calendar year that transition occurred. This continued reimbursement
for the remainder of the calendar year ensured that the participant incurred no financial hardship in
relation to the Medicare Part D plan due to participation in the MHTS.4
Lapse of coverage. Two participants lost Medicare coverage during the study because of a medical
Continuing Disability Review (CDR) that commenced prior to study enrollment. They were able to
enroll in the study because they were appealing the decision to have their coverage revoked and
therefore were still allowed to have Medicare until the final decision was made. When these
participants lost their coverage, the study enrolled them in state-funded high-risk insurance
programs.
Enrolling Uninsured Participants
Westat assisted uninsured participants to enroll in high-risk insurance plans in their state, when one
was available. The WIP sent applications to the site RA via email, which in turn assisted the
participant with the form completion. The WIP then reviewed and submitted completed
applications to the insurance company along with any initial enrollment fees. When the WIP
received enrollment confirmation, the information was documented in the SMS and a payment
schedule was set up.
4
The Westat Institutional Review Board required the study to take precautions to ensure that participants incurred no harm as a result of participation
in the MHTS. Participants were to leave the study no worse off than they were at the start of the study. SSA concurred with this view.
Chapter 7: Health Care and Supported
Employment Financing
7-12
Enrolling Participants in a New Health Plan
Table 7-2 shows the number of participants that Westat staff assisted with enrollment into a health
plan for the first time. The table breaks this information out by year and by type of coverage. The
HBP implementation resulted in new enrollments of 21 participants in Medicare Part B, 50
participants in a Medicare Part D plan, and 3 participants in a private insurance plan (e.g., state highrisk pool). Thus, the WIP assisted enrollment of 74 participants in a new health plan that would
otherwise lack adequate coverage. In addition to acquiring coverage, the WIP assisted a number of
participants switching Part D plans, comparing private, HMO, and Part B plan options, and in
general, making sure they received and maintained adequate health coverage throughout the study.
Table 7-2.
Number of participants enrolled in a new health plan by the WIP, by year and by
type of coverage
Insurance coverage
Medicare Part B
2007
0
2008
13
Medicare Part D
23
25
2
50
Private Insurance
0
2
1
3
23
40
11
74
Total
2009
8
Total
21
Determining Premium Payment Amounts to Participants
As noted earlier, the standard (without late penalties) monthly Part B premium was $93.50 in 2007,
and the monthly premium for a typical Part D plan was approximately $50. Thus, the MHTS
average expected premium payment for a participant who had both Medicare Part B and D was
roughly $143.50 per month. Table 7-3 summarizes the changes in Medicare Part B premiums during
the course of the study.
The MHTS paid the entire premium, up to a predetermined maximum amount (see the following
page) for a participant who had private insurance that covered only him or her. However, when a
participant was not the only individual covered on an insurance plan policy, the study paid only that
portion of the monthly premium that represented the participant’s share. Furthermore, Westat
reimbursed the person identified as the primary insured, (not necessarily the participant), for the
premium or premium portion.
Chapter 7: Health Care and Supported
Employment Financing
7-13
Table 7-3.
1
Medicare Part B premium changes
Year
2006
Medicare Part B
premium
$88.50
Minimum premium
reimbursement amount1
$138.50
2007
$93.50
$143.50
2008
$96.40
$146.40
2009
$96.40
$146.40
2010
$96.50
$146.50
The minimum premium reimbursement amount was always the Medicare Part B premium plus $50 (typical Medicare Part D premium)
for each year.
As a result, Westat set the minimum premium reimbursement amount at $143.50 per month ($93.50
for Part B equivalent coverage and $50.00 for Part D equivalent coverage) for participants with
private insurance coverage for 2007. The minimum premium reimbursement amount increased as
the Medicare Part B premium increased. A maximum premium reimbursement amount of $350 per
month was also set. Actual premium reimbursement amounts for private insurance coverage were as
follows:

If a plan covered the participant and one other person, the study paid 50 percent of the
monthly premium up to the maximum premium reimbursement limit.

If a plan covered the participant and two or more other persons, the study paid 40
percent of the monthly premium up to the maximum premium reimbursement limit.

If the total monthly premium payment for the private insurance was less than $146.50
(in 2010), the study paid the total premium amount.

If the participant’s portion of the monthly premium for the private insurance was less
than $146.50, the study paid $146.50.

A $350 per month, per participant cap was set so that no participant received any
reimbursement above that amount.
For participants with Veterans Affairs (VA) benefits who also had Part B and Part D coverage, the
study reimbursed the premiums associated with those plans. MHTS offered participants covered by
VA benefits but without Part B and Part D coverage enrollment in those plans. The site RA
informed participants about the necessity of Part B and Part D Coverage if they wanted to receive
health services from providers outside the VA system. If a VA-insured participant asked to be
enrolled in Part B and Part D, as with a participant on Medicaid, he or she would be considered to
have ―creditable coverage‖ and therefore could be assisted with enrollment in Parts B and D plans
outside the open enrollment window.
Chapter 7: Health Care and Supported
Employment Financing
7-14
Treatment group participants who were dual-eligible (had both Medicare and Medicaid coverage)
were enrolled in Medicare Part B only if they specifically requested such enrollment. As previously
discussed, Medicaid paid for Medicare premiums for dual-eligible participants. In addition, the study
obtained Part B and Part D coverage and started paying the associated monthly premiums for
participants who lost Medicaid coverage during the period of study participation.
Paying Prescription Medication Co-Pays
One of the study requirements was to cover participant out-of-pocket expenses associated with
study-related services. As discussed elsewhere, a participant could file a claim for any expense they
believed related to their efforts to return to work. Some of these included making co-payments for
behavioral health visits to an off-site provider, transportation to or from a medical appointment, etc.
SSA expected that the study also would cover out-of-pocket expenses associated with the purchase
of prescription medications for the treatment of mental disorders. Even with the promise of
eventual reimbursement through the HBP, some participants would find the initial out-of-pocket
expenses a substantial financial barrier. SSA requested that the investigators explore ways to
overcome this potentially expensive problem for participants seeking treatment. The solution—at
least for prescription co-payments—was a debit card. Subsequently, Westat issued debit cards to all
participants for purposes of making prescription co-payments associated with behavioral health
medications. A more complete description of the debit card plan and implementation appears in the
Supplemental Appendix.
Paying Essential Work-Related Expenses
An additional responsibility of the study was covering essential work-related expenses (EWRE).5 It
is often the case with low-income people with disabilities that the difference between getting a job
and not getting a job is the expense of some incidental item. Some examples are the cost associated
with retaking the nursing licensing exam to update a license, and obtaining appropriate work attire
(special uniforms or interview clothes), and obtaining dental work or prescription lenses. The study
always considered requests for payment for these types of expenses. The Nurse Care Coordinator
was required to complete a special form and submit it to an identified member of the investigative
5
EWRE should not be confused with the SSA support program known as Impairment-Related Work Expenses (IRWE). The EWRE was designed
specifically for the HBP.
Chapter 7: Health Care and Supported
Employment Financing
7-15
team for review and approval. These types of expenses were rare in the study. Approvals of these
types of expenses occurred when a clear and direct link between the item and a specific potential job
was imminent. An example of expenses not approved was general software training (such as
Microsoft suite of general business use software). A few exceptions occurred when an item clearly
improved the participant’s opportunity to find employment. An example that occurred more than
once in the study was the purchase of dentures for participants without teeth.
SE Financing
The MHTS investigators made many decisions in the course of designing the HBP and
implementing its functions. This section details some of the team’s key decisions made with regard
to the provision of SE services.
Decision 1: Replicate the fee-for-service financing environment in SE payment mechanisms.
During the past decade state mental health authorities have moved away from funding psychiatric
rehabilitation services using grants and have moved toward fee-for-service financing using Medicaid
funds. To generate SE expenditure data that are relevant to policymakers, the study initially
proposed to replicate this fee-for-service environment. Likewise, the MHTS required accurate data
on the number and types of participants’ SE and related encounters. To achieve these objectives,
Westat initially required sites to complete and submit Monthly Encounter Forms for each treatment
group participant in order to receive payment for SE services delivered to treatment participants.
Sites submitted forms monthly to ValueOptions, the claims processor for the study. ValueOptions
extracted encounter data from these forms, which allows the study team to construct expenditures
estimates of the SE services delivered. Appendix 7B shows the Monthly Encounter Form.
Decision 2: Recognize the cost to study sites of carrying inactive participants on their SE
caseloads. Westat and the study sites realized that treatment group participants represented a
different type of client than many of the community mental health centers serving as study sites
were accustomed to serving. Typically, clients enrolled in other services at the agency, and
approached the site with a stated desire to work and receive SE services. These clients often remain
on a wait list until an SE ―slot‖ becomes available. In the study, however, sites first reached out to
the participant, who may not have considered the possibility of returning to work. The participant
therefore, may have enrolled in the study with some uncertainty about working or because of the
many incentives treatment group participation offered (e.g., payment of insurance premiums and
out-of-pocket behavioral health care expenditures). In the focus groups with the study site directors,
Chapter 7: Health Care and Supported
Employment Financing
7-16
sites voiced this concern. Site directors were also concerned that participants might fill SE slots
without actively participating in the treatment intervention. If this occurred, sites would be unable to
generate income by billing for services while still having to pay the full salary of an SE specialist (a
fixed cost). Furthermore, as an intent-to-treat study design, the rules forbid dropping or replacing
inactive participants. To cushion any such revenue losses that sites incurred due to participants not
participating in the intervention—and in recognition of the fixed cost component of carrying MHTS
participants on the SE specialist’s caseload—Westat developed an SE payment schedule that paid
sites a minimum monthly amount per participant, regardless of the participant’s level of study
participation.
Decision 3: Base sites’ SE payments on participant insurance coverage, site billing
capabilities, and the number of face-to-face participant contacts. The MHTS initially paid sites
for delivering SE based on three factors:
1.
Whether the site billed Medicaid for SE;
2.
Whether the site billed a third party for other behavioral health services delivered to the
beneficiary; and
3.
The number of face-to-face SE and related encounters in the billing month.
The study paid less to a study site that was able to collect revenue through another payer for other
behavioral health services. The rationale was that the first site collected revenue from both a thirdparty insurer and the study, while the second site depended entirely on the study for payment of the
SE and other behavioral services.
A site’s ability to bill Medicaid—the only insurer that pays any part of SE—depended not only on
the participant having Medicaid coverage, but also on the site’s billing capabilities. As noted earlier,
many study sites were accustomed to receiving grants or capitated payments for their services and
consequently lacked the administrative capacity to bill insurers such as Medicaid or Medicare on a
fee-for-service basis. The MHTS did not penalize sites for their lack of third-party billing expertise.
Instead, the study paid these study sites for SE at the highest rate, as they had no other means of
generating revenue for MHTS participants. Finally, the evidence-based IPS approach to SE
demanded frequent face-to-face client contact. To encourage this contact, SE payments to sites also
depended on the number of face-to-face contacts between the participant and SE specialist (or case
manager providing SE-related case management). In general, the study SE payment rates were to
cover the cost of additional non-face-to-face contacts that were also a part of SE. Appendix 7C,
Original and New Supported Employment and Related Services Payment Schedules, shows the SE
Chapter 7: Health Care and Supported
Employment Financing
7-17
payment schedule used at the start of the study in the columns marked ―Original Schedule.‖ Almost
from the beginning of the study, Westat negotiated alternate payment schedules for 5 of the 22
sites—Maryland, Minnesota, New York, and two sites in Connecticut—because of unique service
and billing circumstances in each of these sites.
Decision 4: Modify the original payment schedule to increase the payment amounts due to
lower than expected revenues to support SE services. The original schedule outlined payments
for participants on Medicaid (Schedule C), participants for whom there was some reimbursement for
other behavioral health services but not for supported employment-related services (Schedule B),
and participants for whom there was no reimbursement for SE-related services or other behavioral
health services (Schedule A). The original schedule paid a minimum of $100 per month for each
participant, even if there was no face-to-face contact.
Sites had raised a variety of concerns about the schedule from its inception, and Westat made a
commitment to the sites to conduct a review of the sites’ experience with the schedule after 6
months. Thus, in July 2007 the MHTS investigators conducted a mid-course review of the study
payment schedule for SE and related services. Overall, Westat needed to determine if the schedule
was serving the study objective of adequately compensating sites for delivering evidence-based
services.
Five related questions emerged as salient for both Westat and the sites:

Is the basic structure of the schedule adequate?

Are the payments adequate to cover study sites’ staff costs?

Does the schedule properly compensate for non-face-to-face work associated with SE
and related services?

Case management often is critical to SE—can the schedule cover the unreimbursed
costs of case management?

Are the rules of the schedule clear to sites?
Westat conducted a fact-finding mission, questioning study site administrators directly to learn more
about these five issues. Westat also interviewed the three Quality Management Project Directors
(QMPDs), experts in the service delivery of IPS SE and evidence-based behavioral health
treatments. The three QMPDs (described in Chapter 2) worked closely with the study sites to ensure
the delivery of high-quality services within the study sites. Therefore, they were familiar with the
Chapter 7: Health Care and Supported
Employment Financing
7-18
structural and financial barriers that sites faced in delivering these services. Westat also reviewed
ValueOptions billing records for SE and other behavioral services in order to document the sites’
actual billing practices in the study’s first 6 months of enrollment. Finally, the investigators
compared the actual participant enrollment data with expected enrollment over this period. This
investigation led to a number of findings, summarized below.
Difficulty Phasing in the SE Specialist
Many of the sites eased into the project by using existing SE specialists while the MHTS participant
caseload grew, rather than hiring new full-time employment specialists to serve a small caseload.
Until the site has nearly 20 participants, this phase-in seemed to be the most efficient strategy.
However, not all sites used this gradual approach to increasing existing employment specialists’
caseloads. Furthermore, some sites had little or no experience with fee-for-service payment
schedules.
Lack of billing. At the time of the review, sites had billed ValueOptions for only slightly more than
half of all enrolled treatment group participants, suggesting that more revenue could come to some
sites if they billed fully. Most of the bills processed by ValueOptions were for only one or two faceto-face contacts each month.
Need for case management. The QMPDs reported that more case management services were
needed for some participants. However, some sites were hesitant to provide these services because
of concern that they were not compensated (for example, case management is not a covered service
under Medicare). However, the Monthly Encounter Form explicitly allowed for the billing of case
management, an opportunity that many sites were missing.
Improper use of the payment schedule. In general, some sites were not using the schedule as
intended and could have generated more revenue had they billed for all participants and for all
eligible face-to-face encounters, including case management.
Unfair compensation for one versus no contacts. QMPDs reported that some sites found it
unreasonable that Schedule B paid the same rate for one contact as for no contacts.
Lack of compensation for non-face-to-face contacts. Many sites were providing considerable
non-face-to-face services involving both employment specialists and case managers. For example,
Chapter 7: Health Care and Supported
Employment Financing
7-19
some participants needed assistance with job development, benefits coordination, housing, and
integration of services. In addition, there were increasing requirements for treatment planning
among clinical staff who did not require face-to-face contact with the participant. These required
services were non-reimbursable.
Low enrollment. Enrollments were lower than expected across all sites. The study paid sites on a
per-participant basis, therefore, lower enrollments limited sites’ revenue-collecting opportunities.
The study originally based study site contracts on an overall enrollment of 3,000 study participants
(1,500 in the treatment group). As discussed in Chapter 2 (Sampling Methodology), the sample was
reduced to a target of 2,200 (1,100 in the treatment group).
Collectively, these obstacles meant that some sites were financially unable to deliver needed
evidence-based services within the MHTS. With respect to the five specific questions articulated
above, the investigators concluded the following:

While the basic structure of the payment schedule remained adequate, there was a need
to pay for one face-to-face contact at a higher rate than for zero face-to-face contacts.
Few sites billed for participants at the top of the schedule (i.e., four or more contacts
per month), but many sites had not implemented SE-related case management services
and when they did, the maximum number of contacts in the schedule (four) may not
have been adequate. The investigators increased face-to-face encounters from five to six
encounters per month to create a proper incentive to deliver these services.

Payment amounts were not adequate within the Supported Employment and Related
Service Payment Schedule to cover study sites’ staff costs due to increased service
demands at each site, particularly involving non-face-to-face activity. This situation
suggested consideration for raising current payment rates substantially.

The schedule did not properly compensate sites for the non-face-to-face SE-related
contacts. Sites should receive a higher payment when there are no face-to-face contacts
with a participant in a given payment month.

Case management is often critical to SE success, and should be encouraged. When
performed face-to-face, the staff should count it as SE-related contacts within the
schedule on the Monthly Encounter Form.

The rules of the payment schedule were not clear to all sites, particularly regarding the
inclusion of case management on the Monthly Encounter Form. Sites required more
technical assistance on billing and its structure and rationale.
Based on the findings from this review, Westat decided to increase the payment amounts
substantially in the schedule because of lower than expected revenues from participant billings, some
Chapter 7: Health Care and Supported
Employment Financing
7-20
of which was due to low enrollment and some to problems in engaging participants in face-to-face
encounters. Appendix 7C compares the original and new payment rates for all three schedules A, B,
and C. All payment schedule changes took effect on August 1, 2007. Though payment amounts
increased dramatically, the same basic structure and reimbursement rules remained in place. The
team considered moving to a grant or cost-based contracting arrangement, but the investigators
decided to maintain the payment structure already in place allowing the study to continue to yield
information about fee-for-service payment schedules. The specific payment schedule changes were
as follows:

Add a payment step to Schedules A and B to differentiate payments for no contacts and
one contact. The team increased payment for no contacts to $200 in all three schedules.

The team further revised the schedule to increase payments at all levels. Schedule C
increased to $200 for all contact levels, as Medicaid billing for all face-to-face services
occurred.

Provided clarification on the rules for coverage of case management in the schedules,
where some outside payment system did not reimburse case management. Identified
payment schedule coverage for case management, which promotes better SE and
integrated care. Provided more technical assistance to the sites about use of the
schedules to pay for SE and case management.
Decision 5: Modify the payment schedule to a “negotiated fixed rate” per participant month
of enrollment. As mentioned, during the review that resulted in Decision 4, the team considered
moving to a grant or cost-based contracting arrangement. However, the team decided to maintain
the existing fee-for-service payment structure. The team was reluctant to abandon the payment
system because it was closest to the usual Medicaid mechanisms for paying for SE. In large part, this
reluctance was due to the expectation that the MHTS experience might influence other health
insurance arrangements to consider covering SE on a fee-for-service basis. The team hoped that the
reporting of specific services as a part of the request for payment would improve the accuracy of
data on service utilization for analysis in the study. However, a second review conducted a few
months after implementing the fee schedule described in Decision 4, confirmed that it was necessary
to modify the schedule, yet again. The decision was to move to a negotiated fixed rate per
participant month of enrollment regardless of the number of face-to-face contacts. In December
2007, Westat began the process of establishing a negotiated fixed rate. The feedback from the site
directors was that the modified schedule (Decision 4) did not sufficiently provide revenue to fund
evidence-based SE and other behavioral health services as required by the study. In addition to the
Chapter 7: Health Care and Supported
Employment Financing
7-21
findings summarized for the rationale for Decision 4, the following factors provided the rationale to
move forward with a negotiated fixed-rate payment schedule:

The provision of evidence-based SE services required ongoing supervision, planning
meetings, and other support for the employment specialists.

Treatment group participants required case management and other non-billable services.

Enrollment rates continued to be significantly less than expected, which limited revenue
but still required full-time employment specialists and case managers.
There continued to be a reasonable number of unengaged participants who did not meet with the
employment specialists face to face. Nonetheless, the investigators expected the employment
specialists to actively attempt to engage or re-engage these participants while the site received the
lowest payment schedule for these efforts.
The continued financial challenges that the sites faced in providing evidence-based services, led to
the site-specific negotiated fixed monthly rate per participant per month enrolled. The final
negotiated flat rates were $400 (12 sites), $450 (6 sites), and $500 (4 sites) per participant per month
enrolled.
HBP Expenditures
This section presents the expenditures associated with the implementation of the HBP. The HBP
spent $14,219,241 for the 1,121 treatment participants in the MHTS, representing $6,342 per person
per year. Table 7-4 shows the distribution of expenditures for all the treatment participants
(n=1,121) and for 24-month treatment participants (n=982). The total expenditure for the 9826
treatment participants enrolled for 24 months was $13,719,519 (representing $6,986 per person per
year), a difference of $499,722. The HBP spent $10,000 or less on about one quarter of the
participants, accounting for $1,749,595 for all treatment participants and $1,335,466 for just the
treatment participants enrolled for 24 months. The HBP spent between $10,001 and $15,000 per
person for about half of the participants during the 24 months of their participation. That accounted
for $6,898,139 for all treatment participants and $6,849,130 for just the treatment participants
enrolled for 24 months. Spending for participants who had expenses between $15,001 and $20,000
accounted for $3,458,374 for all treatment beneficiaries and $3,442,355 for just the treatment
6Excluding
withdrawals, administrative drops, and deaths.
Chapter 7: Health Care and Supported
Employment Financing
7-22
participants enrolled for 24 months. The original contract placed a $10,000 per person annual limit
on expenditures, but the MHTS investigators renegotiated this limitation as an average annual
expenditure. The modal expenditure was considerably less than the original $10,000 annual limit.
However, 78 individuals had expenditures of $20,001 or more, accounting for $2,113,133 among all
treatment participants, and for the 77 individuals of the 78 of them in this range (among those
enrolled for 24 months) accounted for $2,092,568.
Table 7-4.
Distribution of HBP expenditures among all participants and 24-month participants
All participants
Total
expenditures
≤ $10,000
24-month participants
(n=1,121)
Total
Frequency
expenditures
285
$1,749,595
(n=982)
Total
Frequency
expenditures
152
$1,335,466
Difference
$414,129
$10,001-$15,000
554
$6,898,139
550
$6,849,130
$49,009
$15,001-$20,000
204
$3,458,374
203
$3,442,355
$16,019
78
$2,113,133
77
$2,092,568
$20,565
1,121
$14,219,241
982
$13,719,519
$499,722
≥ $20,001
Total
Figure 7-1 illustrates the components of the total of $14,219,241 in HBP expenditures. SE was farand-away the largest area of spending for treatment participants, accounting for approximately 72
percent of expenditures or $10,227,400. Of course, most often there were no other payment sources
for SE services other than the HBP. The next largest areas of expenditure were for health insurance
premiums (11%), followed by behavioral health spending (7%), and medication expenditures from
the debit card arrangement (8%). Spending was one percent each for employment-related work
expenses and transportation, miscellaneous expenditures to participants for research participation,
and general medical expenses.
Chapter 7: Health Care and Supported
Employment Financing
7-23
Figure 7-1.
Distribution of total expenditures by type of claim including all participants
(n=1,121) (based on date of service)
Unclassified
1%
Supported
employment
claims
72%
Behavioral health
services
7% Prescriptions EWRE and
8%
transport
1%
General medical
<1%
Premium
11%
Figure 7-2 illustrates the components of the total of $13,719,518 in HBP expenditures for the 24month participants. SE was again the largest area of spending for 24-month treatment participants,
accounting for approximately 71 percent of expenditures or $9,800,225. The next largest area of
expenditure was again health insurance premiums (12%), followed by prescription medications (8%),
and behavioral health spending (7%). Spending was one percent or less each for employment-related
work expenses and transportation, miscellaneous (unclassified) expenditures to participants for
research participation, and general medical services.
Figures 7-3 and 7-4 illustrate the implementation expenditures of the HBP over time. Figure 7-3
shows the implementation in the life of the project and demonstrates that per person per quarter
expenditures reached a steady state of approximately $600 per month by March 2008. Spending
started out at approximately $400 per person per month and increased slowly over time until
reaching the steady state. The rise coincides with the two changes in payment for SE described in
the earlier sections of this report. Since SE spending accounts for 72 percent of the total, these
changes in payment rates drive the implementation spending observations. There is one additional
peak in expenditures to approximately $650 per person per month in the quarter from April – June
2008, coinciding with the renewal payments of annual insurance premiums.
Chapter 7: Health Care and Supported
Employment Financing
7-24
Figure 7-2.
Distribution of cumulative expenditures by type of claim including all 24-month
participants (n=982) (based on date of service)
Unclassified
1%
Behavioral health
services
7%
Prescriptions
EWRE and
8%
transport
1%
General medical
<1%
Premium
12%
Supported
employment
claims
71%
Average HBP expenditures per person-per quarter (n=1,121)
Average spending per participant
Figure 7-3.
Chapter 7: Health Care and Supported
Employment Financing
7-25
Average HBP expenditures per participant month (n=1,121)
Average payment per FTE
Figure 7-4.
Figure 7-4 illustrates the spending pattern for each month of individual participation in the 24
months of the intervention. It presents the monthly payments on behalf of treatment participants
for their 24 months of potential tenure in MHTS. As would be expected, the amount of spending is
virtually identical to the per person-per month spending shown in Figure 7-3—with a steady state of
approximately $600 per person per month. The figure also shows the initial spending for insurance
premiums in addition to the other spending, spread out throughout the period. Per person-per
month spending is almost $700 in the first month, and the peak to approximately $650 in month 8-9
is for the renewal of the insurance premiums for the second year of participation.
Discussion
For the most part, expenditures within the HBP conformed to expectations about financing MHTS
services. SE dominated the intervention and accounted for slightly more than 70 percent of HBP
expenditures. While the demonstration initially intended to pay for SE on a fee-for-service basis, the
project decided to negotiate a monthly case rate per person enrolled to cover the cost of
employment specialists, as well as some case manager time and supervision by senior clinical staff.
As a result, the SE payments ranged from approximately $4,800 to $6,000 per enrolled participant
per year. Financing with case rates provides steady payments to sites, as reflected in Tables 7-3 and
7-4, above, once the final payment rates were negotiated. Payment for health insurance premiums
accounted for slightly more than 10 percent of spending and out-of-pocket health expenditures, paid
Chapter 7: Health Care and Supported
Employment Financing
7-26
directly or using the debit card for medications, represented together about 15 percent of
expenditures. Covering health insurance premiums was a unique element of the MHTS that made
participation in the demonstration particularly advantageous for treatment group participants.
Reimbursing for out-of-pocket expenses meant that treatment participants had virtually no cost to
themselves for participation in the MHTS. This was one of the objectives of the HBP, and it was
implemented successfully.
Lessons from implementing the HBP. In the sections of this chapter prior to the discussion of
expenditures, the lessons from the HBP experience are outlined as a series of critical decisions in
implementation. These decisions could guide any future effort to finance a complex set of services,
similar to those involved in the MHTS, as well as to reimburse participants for the health insurance
premiums and out-of-pocket costs. The MHTS experience suggests that most individuals attracted
to a demonstration like the MHTS have health insurance and do not need much assistance in
obtaining appropriate coverage, but an important minority do need assistance. The MHTS
developed a HBP that successfully matched participant needs with appropriate Medicare coverage,
particularly with the then-new Part D prescription drug plans. The HBP also successfully
implemented a debit card arrangement to prevent out-of-pocket costs for medications rather than
retrospectively reimburse them, another benefit of MHTS participation. The HBP developed and
tailored an approach to financing SE that could be emulated in any future demonstration. The
experience suggests that each site will have its own special needs for financing and that a standard
fee-for-service arrangement, even with complex payment rates, is unlikely to work well.
Furthermore, it is important to recognize that sites that participate in such a demonstration will have
upfront costs to start up service, will have costs associated with outreach to all individuals who are
enrolled even if they do not use SE services, and that case management and supervision are ―hidden
costs‖ of participation for any site.
The design of the MHTS does not permit an exact assessment of how essential the HBP was to
enrollment or to subsequent success of the overall intervention. There were many elements to the
MHTS, and no single one can be isolated and assessed for its specific impact. It is clear, however,
that obtaining payment for all behavioral health care costs and SE, along with payment of insurance
premiums, removed some of the putative barriers to participation in active efforts to return to work
for individuals with a disabling mental disorder on the SSDI rolls. The experience of the MHTS
suggests that it is feasible to implement a payment system to cover these services without any outof-pocket expenditure for SE or related behavioral health services, including medications and
psychosocial services. Though the negotiated payment rate per site, per participant was between
$400 and $500, the expenditures for all but 78 of 1,121 participants did not exceed the original
Chapter 7: Health Care and Supported
Employment Financing
7-27
expected limit of $10,000 annually, and the modal payment was far less than that amount. These SErelated expenditures accounted for slightly more than 70 percent of all HBP expenditures.
If an agency such as SSA wanted to implement the MHTS on a broader scale, the MHTS experience
describes the expected spending of a HBP to pay for SE and also to insulate participants from other
out-of-pocket expenditures associated with IPS SE, behavioral health treatment, and related efforts
to return to work. Access to evidence-based services was a hallmark of the MHTS. The HBP made
it possible to provide such access with virtually no out-of-pocket costs to participants.
Chapter 7: Health Care and Supported
Employment Financing
7-28
Chapter 8
Utilization of Services
This chapter reports on the utilization of healthcare services of MHTS study participants in the
treatment and control groups. Impacts of the MHTS intervention on physical and mental health
service utilization levels and patterns are potentially important from a budgetary and social cost
standpoint. There is a substantially reduced net cost to government if the intervention reduces
overall utilization of these services, given that other public programs finance the bulk of these
services. Funds for these services come primarily from Medicare and Medicaid, but also from other
public programs such as care from the Department of Veterans Affairs (VA), state vocational
rehabilitation, or other agencies.
Service utilization data collected from treatment and control group participants during their 24months of participation in the MHTS provide the basis for constructing estimates of the impact of
the treatment intervention on service use. The types and measures of utilization studied include
number of emergency room (ER) visits, number of hospital admissions, number of nights spent in
the hospital, number of visits to receive outpatient psychiatric emergency or crisis services, and
number of visits to other mental health providers (e.g., clinics or therapists). With the intent-to-treat
model, the primary focus is on unconditional mean values for these measures comparing participant
service use in the treatment group to that in the control group.
The primary source of data for these analyses of participant service utilization in the MHTS came
from the eight post-baseline computer-assisted interviews with each study participant. The
interviews included a section on health care service utilization, asking participants to recall details
about ER visits, ER visits that resulted in an inpatient (overnight) hospital admission, other
overnight hospital stays, psychiatric emergency or crisis (outpatient) visits, and other clinic or mental
health provider visits. “Other” clinic or health care provider visits included ongoing (regularly
scheduled) visits to mental health providers such as prescribers and therapists. During each postbaseline interview, participants recalled these events since the date of the last completed interview,
which allowed for collection of complete data on service utilization even if the participant skipped a
post-baseline interview. For instance, if participants skipped the previous interview (i.e., 3 months
ago), they were asked about service utilization in the past 6 months, as opposed to the typical recall
period of 3 months.
Chapter 8: Utilization of Services
8-1
Dependent Variables
The data collected in the computer-assisted interviews required manipulation to create useful
analytic variables to assess the extent to which participants used various healthcare services during
the 24-month study period. The descriptions below include each of the specific service utilization
variables included in this chapter.
ER visits. During each interview, participants responded to questions about visits to an emergency
room (since the date of the last completed interview). Interviewers requested the participant
enumerate each ER visit, and then for each ER visit, report the reason for the visit and if the visit
resulted in a hospital admission. A constructed group of variables included the number of ER visits
a participant made for any given month of the 24-month study period, specifically for a mental
health problem, or for some other problem.
Outpatient visits for psychiatric emergency or crisis. Interviewers asked participants if they
received help for a psychiatric emergency or crisis from some source other than an emergency room
or hospital. When appropriate, they enumerated each event and listed all providers associated with
the event. In addition, interviewers requested information about the outcome of the visit. These data
allowed for construction of a series of variables indicating the number of outpatient psychiatric
emergency or crisis visits since the date of the last completed interview. When combined, the
interviews provide a complete picture of outpatient psychiatric emergency or crisis visits over the
entire length of study participation. Participants also provided an outcome for each visit, including
(1) obtained a prescription or consult on medication, (2) mental health counseling, or (3) some other
outcome (i.e., vocational counseling, spiritual or religious counseling, or peer support).
Other mental health provider visits. After asking about ER visits, hospital admissions, and
outpatient visits for psychiatric emergencies or crises, the interviewer requested information about
care received from any other clinic or mental health provider (since the date of the last interview).
For all clinics or providers visited, participants responded to questions about the reason for the visit
(i.e., physical problem, mental health problem, alcohol problem, drug problem, or some other
problem), how many times they went to the facility, and the names of all providers seen at the
facility. Finally, the interviewer requested information about the outcome of each visit. Each visit
included an appropriate outcome code as noted above.
Hospital admissions. For each ER visit reported, participants were asked if the ER visit resulted in
a hospital admission. If so, participants responded to questions about the reason for the visit (i.e.,
Chapter 8: Utilization of Services
8-2
physical problem, mental health problem, alcohol problem, drug problem, or some other problem)
and the number of nights stayed in the hospital. In addition to hospital stays following an ER visit,
participants reported any other hospital stays since the date of the last completed interview (i.e., any
other hospital stays not already mentioned during the interview). Again, for each hospital stay
reported, participants responded to questions about the reason for the visit and the length of the
stay. These data allowed for the construction of a series of variables indicating the number of
hospital admissions, the number of hospital nights, and the reason (mental health problem or for
some other problem, including physical health) for the hospitalization. The resulting statistics
included the mean number of hospital admissions for each participant during the study period for
mental health problems and for other problems (not related to mental health), and the mean number
of nights stayed in the hospital (for mental health problems or for other problems not related to
mental health).
The service utilization data from the interviews required a small number of minor adjustments to
account for reporting anomalies or missing data. In a very small number of cases, respondents who
reported an overnight stay in the hospital also reported either (a) that the number of nights stayed
was zero or (b) that they did not know the number of nights they stayed. In the former case (i.e.,
zero nights stayed), the hospital stay was excluded from our analyses. In the latter case, the number
of nights stayed was imputed either from data on other hospital stays for the same respondent or
from data on average nights stayed as reported by other respondents.
On rare occasions, the participant responded, “don’t know” or refused to answer some of the
service utilization items. Such recorded responses were set to missing and dropped from the
analysis. In other instances (also rare), the respondents indicated that a service use had occurred but
when asked how many times the service was used, the recorded response was either “don’t know”
or “refused.” In these cases, imputations filled in the number from data on other respondents.1
In summary, this chapter provides estimates of the MHTS effects on the following principal
measures of service utilization:
1

Number of overnight hospital stays,

Number of days of overnight hospital stays,
Note that this imputation step applied only to numbers of visits to psychiatric crisis or emergency centers (excluding hospitals or emergency rooms)
and to numbers of visits to other clinics or mental health providers. This imputation step was not relevant for numbers of overnight hospital stays or
ER visits since each hospital stay was reported separately; so no participant was asked how many stays he or she had (though the participant was
asked for each stay how many nights he or she stayed), and each ER visit was also reported separately.
Chapter 8: Utilization of Services
8-3

Number of ER visits,

Number of visits to psychiatric crisis or emergency centers (excluding hospitals or
emergency rooms), and

Number of visits to other clinics or mental health providers.
The analysis includes estimation of MHTS effects for the full two-year followup period, for the last
year of the followup period, and for the first year of the followup period. Each analysis uses the five
principal service use measures (regardless of the reasons for the service use), and for three additional
measures of hospital stays, hospital days, and ER visits for reasons relating to mental health
problems. For respondents who did not complete the fourth quarterly followup interview, service
use data for the first year were computed by allocating service use in the next subsequent completed
followup evenly across the quarters of the recall period for that followup interview.
Covariates
The number of potentially relevant covariates in the data is very large, and even with over 2,000
observations, to include all of these potentially relevant covariates in regression models of MHTS
effects could present major problems of over-fitting and multicollinearity. For these reasons, a
limited number of measures and patterns of service use during the baseline period served as useful
controls for: (1) multiple dimensions of baseline health status, (2) the presence of baseline health
problems, and (3) the risk of future (post-baseline) problems that result in mental health or somatic
service use. Controlling for the volume and pattern of baseline service use suggests that the omission
of other covariates would not bias the results and would not substantially reduce the efficiency of
the MHTS effect coefficient estimates.
Estimation Models
The rationale for the selected estimation method and models was based primarily on the
distributional characteristics of the dependent variables, in particular, the fact that (1) they are all
count (integer) data, and that (2) they are characterized by a large mass of observations at zero (i.e.,
no visits for the particular variable). A regression model developed specifically for analyzing such
data is the zero-inflated negative binomial regression. Estimation of this regression model (using
Stata) also provides straightforward tests of two related models, namely, a zero-inflated Poisson
regression and a standard negative binomial regression (without inflation for excess zeros). Specific
Chapter 8: Utilization of Services
8-4
coefficient values were obtained via maximum pseudo-likelihood estimation. All models, except
those with site-specific dummy variables, used an estimated variance-covariance matrix for random
errors that allowed for clustering (i.e., correlation) of such errors across individual respondents from
the same site.2
In the models reported below, estimates of MHTS effects include four different specifications of
covariates: (1) no covariates, (2) the baseline value of the relevant dependent variable as the only
covariate, (3) baseline values for all five primary service use variables, and (4) baseline values for all
five primary service use variables plus site-specific “dummy” variables.
Results
Descriptive statistics. Table 8-1 gives the mean values for all dependent variables and all covariates
based on overall service utilization measures for the total sample and for the treatment and control
groups separately. Note that the computed means used the post-stratification weights and assigned
zero weights to persons designated as non-respondents (defined in Chapter 2). Differences in
baseline values of service utilization between treatment and control groups were generally very small.
For two of the baseline measures (psychiatric crisis or emergency center visits and days of overnight
hospital stays) the mean values for the treatment group were about 20 percent below those for the
control group; however, for the three remaining baseline measures, the treatment and control group
means were almost equal. It is interesting to note that, after accounting for the differences in
duration between the 24-month followup period and the baseline recall periods, mean rates of
service use were not dramatically different between the Baseline and Followup interviews. The
exceptions to this were: (1) the decline in rates of visits to psychiatric crisis or emergency centers in
the followup period, (2) the modest increase in rates of overnight hospital stays in the followup
period, and (3) the large increase in rates of visits to other clinics or mental health providers in the
followup period.
2
When site-specific dummies were included instead of error clustering by site, coefficient estimates were obtained via maximum likelihood.
Chapter 8: Utilization of Services
8-5
Table 8-1.
Weighted mean values and N's for treatment and control groups and total: Overall
utilization measures
Variable Definition
DEPENDENT VARIABLES
Overnight hospital stays
Overnight hospital stays – year 1
Overnight hospital stays – year 2
Days of overnight hospital stays
Days of overnight hospital stays –
year 1
Days of overnight hospital stays –
year 2
Emergency room visits
Emergency room visits – year 1
Emergency room visits – year 2
Outpatient psychiatric crisis or
emergency center visits
Outpatient psychiatric crisis or
emergency center visits – year
1
Outpatient psychiatric crisis or
emergency center visits – year
2
Other clinic or mental health
provider visits
Other clinic or mental health
provider visits – year 1
Other clinic or mental health
provider visits – year 2
BASELINE COVARIATES
Overnight hospital stays – past 12
months
Days of overnight hospital stays –
past 12 months
Emergency room visits – past 6
months
Outpatient psychiatric crisis or
emergency center visits – past
6 months
Other clinic or mental health
provider visits – past 3 months
Treatment
Weighted
Mean
N*
Control
Weighted
Mean
N*
Total
Weighted
Mean
N*
0.86
0.45
0.42
5.74
3.21
892
1,006
897
892
1,006
0.97
0.51
0.47
7.37
4.16
973
1,057
977
973
1,057
0.92
0.48
0.45
6.57
3.69
1,865
2,063
1,874
1,865
2,063
2.60
897
3.25
977
2.93
1,874
1.91
0.95
0.95
0.52
890
1,008
895
896
1.94
0.97
0.99
0.90
968
1,056
974
978
1.92
0.96
0.97
0.72
1,858
2,064
1,869
1,874
0.41
1,009
0.56
1,062
0.49
2,071
0.15
899
0.31
980
0.23
1,879
51.54
892
35.34
970
43.29
1,862
24.87
1,010
17.90
1,059
21.39
2,069
26.24
895
17.14
975
21.60
1,870
0.38
1,015
0.39
1,060
0.38
2,075
2.87
1,015
3.53
1,060
3.20
2,075
0.46
1,015
0.46
1,061
0.46
2,076
0.31
1,015
0.39
1,063
0.35
2,078
5.26
1,014
5.28
1,064
5.27
2,078
* N is the unweighted number of respondents with a weight > or = 1. It excludes respondents with weight = 0 and other respondents
whose variable values were missing.
Table 8-2 shows the analogous mean values for dependent variables and baseline covariates based
on additional mental-health-specific service utilization measures. Treatment versus control group
baseline differences are somewhat larger here, with the treatment group having lower values of
Chapter 8: Utilization of Services
8-6
hospital stays (0.16 vs. 0.21; 25% lower), hospital days (1.67 vs. 2.62; 36% lower), and ER visits (0.12
vs. 0.13; 8% lower).
Table 8-2.
Weighted mean values and N's for treatment and control groups and total: ER and
inpatient mental health utilization measures
Variable Definition
DEPENDENT VARIABLES
Mental health overnight hospital
stays
Mental health overnight hospital
stays – year 1
Mental health overnight hospital
stays – year 2
Days of mental health overnight
hospital stays
Days of mental health overnight
hospital stays – year 1
Days of mental health overnight
hospital stays – year 2
Mental health emergency room
visits
Mental health emergency room
visits – year 1
Mental health emergency room
visits – year 2
BASELINE COVARIATES
Mental health overnight hospital
stays – past 12 months
Days of mental health overnight
hospital stays – past 12 months
Mental health emergency room
visits – past 6 months
Treatment
Weighted
Mean
N*
Control
Weighted
Mean
N*
Total
Weighted
Mean
N*
0.40
892
0.45
973
0.42
1,865
0.21
1,006
0.26
1,057
0.23
2,063
0.19
897
0.20
977
0.20
1,874
3.86
892
4.92
973
4.40
1,865
2.19
1,006
2.98
1,057
2.58
2,063
1.70
897
2.01
977
1.86
1,874
0.39
890
0.52
968
0.45
1,858
0.20
1,008
0.27
1,056
0.23
2,064
0.19
895
0.26
974
0.22
1,869
0.16
1,015
0.21
1,060
0.19
2,075
1.67
1,015
2.62
1,060
2.14
2,075
0.12
1,015
0.13
1,061
0.12
2,076
* N is the unweighted number of respondents with a weight > or = 1. It excludes respondents with weight = 0 and other respondents
whose variable values were missing.
Regression results for estimated MHTS effects. Table 8-3 shows the estimated treatment
effects, based on the zero-inflated negative binomial regression coefficients, for the MHTS
treatment dummy variable. Note that these treatment effects are the average differences between
predicted values of the dependent variables (for values of zero vs. one for the treatment dummy)
where the average is computed over the entire weighted study sample of participants.
Chapter 8: Utilization of Services
8-7
Table 8-3.
Estimated average marginal treatment effects
Dependent Variable
A. Emergency room visits
None
Baseline emergency room visits
All 5 baseline overall use measures*
All 5 baseline overall use measures,
site dummies**
B. Outpatient psychiatric crisis visits
None
Baseline emergency room visits
All 5 baseline overall use measures*
All 5 baseline overall use measures,
site dummies**
C. Other clinic or mental health provider
visits
None
Baseline emergency room visits
All 5 baseline overall use measures*
All 5 baseline overall use measures,
site dummies**
D. Overnight stays
None
Baseline emergency room visits
All 5 baseline overall use measures*
All 5 baseline overall use measures,
site dummies**
E. Overnight days
None
Baseline emergency room visits
All 5 baseline overall use measures*
All 5 baseline overall use measures,
site dummies**
Year 1
Est.
Effect
p-value
Year 2
Est.
Effect
p-value
2-Year Estimate
Est.
Effect
p-value
-0.02
-0.07
-0.06
-0.06
0.788
0.326
0.404
0.379
-0.04
-0.06
-0.06
-0.05
0.464
0.247
0.248
0.430
-0.04
-0.06
-0.06
-0.05
0.464
0.247
0.248
0.430
-0.15
-0.15
-0.14
-0.21
0.164
0.146
0.203
0.166
-0.17
-0.24
-0.25
†
0.311
0.043
0.096
†
-0.39
-0.45
-0.48
-0.74
0.019
0.004
0.037
0.006
6.97
8.88
9.09
8.40
<0.001
<0.001
<0.001
<0.001
9.10
9.88
9.99
8.79
0.004
0.003
0.002
<0.001
16.20
19.57
19.87
17.56
0.001
0.001
<0.001
<0.001
-0.06
-0.05
-0.05
-0.05
0.130
0.194
0.200
0.268
-0.05
-0.06
-0.07
-0.09
0.233
0.161
0.13
0.038
-0.12
-0.13
-0.14
-0.16
0.092
0.041
0.035
0.024
-0.95
-0.71
-0.78
-0.87
0.016
0.064
0.049
0.061
-0.65
-0.75
-0.87
-1.14
0.064
0.082
0.035
0.012
-1.63
-1.63
-1.75
-2.29
0.008
0.025
0.006
0.003
*Baseline value of outpatient psychiatric crisis visits is excluded from the model for Year 1.
**Other clinic or mental health visits baseline and outpatient psychiatric crisis visits baseline are excluded for Year 1; outpatient
psychiatric crisis visits baseline are excluded for Year 2.
†Not applicable. Maximum likelihood estimation did not converge.
The following analytic reports for each of the overall service utilization measures include estimates
of MHTS effects for (1) the full 24-month period, (2) the first 12 months post-baseline, and (3) the
second 12 months post-baseline. Four different specifications for the regression covariates comprise
the estimates, including:
1.
Regressions using only the MHTS treatment dummy and no baseline covariates,
2.
Regressions using the MHTS dummy and the baseline value for the dependent variable,
Chapter 8: Utilization of Services
8-8
3.
Regressions that add the baseline values for the four additional overall utilization
measures to specification (2), and
4.
Regressions that use specification (3) but also include site-specific variables.
Reported p-values for MHTS effects based on specifications (1), (2), and (3) also take into account
clustering of error terms among observations for participants in the same site; specification (4) does
not account for site-specific clustering since it includes site-specific variables.
Looking first at the estimates of MHTS effect on use of ER visits (Table 8-3, panel A), only weak
and clearly non-significant negative effects were observed. While the inclusion of baseline ER visits
tends to increase the absolute size of the estimated impact (relative to specification (1) with no
covariates), results are in general qualitatively and quantitatively similar across all four specifications.
In contrast, estimated MHTS effects on numbers of psychiatric crisis or emergency center visits
(Table 8-3, panel B) for the full 24-month followup are strongly negative in all specifications. It is
also interesting to note that the magnitudes of these effects (approximately 0.5 visits less) are large
relative to the mean observed rates of 0.72 visits for the 24-month followup. Analogous results for
the first 12 months of followup show MHTS effects that are consistently negative, clearly not as
significant as the 24-month results, and a decline in magnitude that is more than proportionate to
the difference in followup periods. Comparing results for the second 12 months of followup with
the first 12 months reveals somewhat stronger and more significant negative MHTS effects in the
former period, especially when baseline covariates are included in the model. This trend suggests
that the impact of the intervention was increasing over time. Note also that for all three followup
periods, as in the case of the ER visit regressions, controlling for baseline covariates and including
site-specific variables increased the magnitude and significance of the estimated MHTS effects.
Panel C of Table 8-3 shows estimates of MHTS effects on visits to other clinics and outpatient
mental health providers. The results are in sharp contrast to the negative and sometimes
insignificant estimates for ER visits, and the negative and significant impacts on psychiatric
emergency visits. Instead, this panel reveals relatively large and highly significant positive impacts of
the MHTS on numbers of visits. It is reasonable to attribute most of these positive impacts to the
design of the treatment intervention itself. The increases in insurance coverage for behavioral health
and the provision of systematic medication management services presumably encouraged treatment
group participants to make ongoing use of behavioral health services. (This is also consistent with a
decline in the need for psychiatric emergency care.) Individual Placement and Support (IPS)
supported employment services in the MHTS were most often provided by a program that was
Chapter 8: Utilization of Services
8-9
integrated with a provider organization that also provided a range of non-vocational behavioral
health services. It is also interesting to note that the magnitude of the MHTS effect on these visits
did not substantially increase from the first 12 months of followup to the second 12 months.
Table 8-3, panels D and E, report estimates of MHTS average effects on overnight hospital use.
Regressions on numbers of overnight stays show consistently negative and significant (p-value < 0.1)
estimates of MHTS effects for the full 24-month followup period. Estimates for the first 12-month
and second 12-month followup periods are also negative but are not significant except for the latter
period when site-specific dummies are included. There is only weak evidence of a tendency for the
negative MHTS effect to increase over time. In addition, the pattern of larger negative impacts when
baseline covariates and site-specific dummies are controlled for is once again observed, but only for
the 24-month and second 12-month followup periods.
Table 8-4 provides estimates of MHTS effects on the additional mental-health-specific utilization
measures (for ER visits, overnight stays, and days in overnight stays). As was true for the estimates
for all ER visits, all overnight stays, and days in all overnight stays, MHTS estimated impacts are all
negative; the main differences in Table 8-4 compared to the results in Table 8-3 pertain to p-values
and magnitudes of the estimated effects. For mental health overnight stays, estimated impacts are
smaller than for all stays (in Table 8-4) and are uniformly not significant. For mental health
overnight days, impact estimates are also smaller and less significant than the corresponding
estimates for all days in Table 8-3, but the impact estimates for two years and for the first year of the
followup has p-values that in several cases are significant or approach significance. Finally, while the
impact estimates in Table 8-3 for all ER visits were clearly and uniformly not significant, the
corresponding estimates in Table 8-4 are all clearly significant and generally larger in magnitude.
Evidence of negative MHTS effects on days of inpatient care in panel B of Table 8-4 is fairly strong
and consistent. Clearly, the tendency for the size of this effect increases with inclusion of baseline
covariates and site dummies but the pattern of increasingly large MHTS effects over time is not
clear. The magnitude of the estimates is also worth noting; they imply that the MHTS effects
(approximately 0.8 to 1 day less of use per year) are just slightly less than one-third the size of the
mean 12-month use figures for baseline and for the first 12-month followup.
Chapter 8: Utilization of Services
8-10
Table 8-4.
Estimated average marginal treatment effects for emergency room visits and
overnight hospital use for mental health problems
Dependent Variable
A. Overnight mental health stays
None
Baseline overnight mental health
stays
All 5 baseline overall use measures*
All 5 baseline overall use measures,
site dummies*
B. Overnight mental health days
None
Baseline overnight mental health
stays
All 5 baseline overall use measures *
All 5 baseline overall use measures,
site dummies*
C. Emergency room visits for mental
health problems
None
Baseline overnight mental health
stays
All 5 baseline overall use measures *
All 5 baseline overall use measures,
site dummies*
Year 1
Est.
Effect
p-value
Year 2
Est.
Effect
p-value
2-Year Estimate
Est.
Effect
p-value
-0.04
-0.02
0.242
0.475
-0.01
-0.003
0.728
0.913
-0.05
-0.03
0.316
0.546
-0.01
-0.01
0.610
0.841
-0.01
†
0.656
†
-0.03
-0.01
0.593
0.789
-0.79
-0.44
0.037
0.171
-0.31
-0.19
0.360
0.570
-1.05
-0.60
0.101
0.301
-0.59
-0.56
0.103
0.182
-0.44
†
0.222
†
-0.93
-1.33
0.144
0.045
-0.07
-0.06
0.048
0.026
-0.07
-0.08
0.032
0.004
-0.13
-0.12
0.026
0.006
-0.07
-0.07
0.004
0.014
-0.08
†
0.004
†
-0.15
†
<0.001
†
*Baseline value of the dependent variable is also included as a covariate.
†Not applicable. Maximum likelihood estimation did not converge.
Results of specification tests. Further analyses examined evidence of two different types of
specification tests. First, evidence testing the null hypothesis that the correct specification is actually
Poisson rather than negative binomial is provided by the estimated confidence intervals on the overdispersion parameter (“alpha”) provided in the regression output (using Stata). In all cases of the
analyses, the results permit rejection of the Poisson specification. Second, the Vuong test provides a
test for the null hypothesis that the correct specification is the negative binomial model rather than
the zero-inflated negative binomial model. Since an algorithm for performing this test is not
available when there is clustering of errors, it was applied only for those models including site
dummies to control for such clustering (rather than using a “vce [cluster]” command in Stata). Once
again, the results of the test consistently rejected the null hypothesis of no zero inflation, thereby
supporting use of the zero-inflated model.
Examining attrition bias: Comparing late dropouts versus non-dropouts. A more limited
analytic approach to assessing the effects of attrition bias compared (a) participants with non-
Chapter 8: Utilization of Services
8-11
missing values for the dependent variables in both the first and second 12-month followup periods
to (b) participants with non-missing values for the first 12-month followup period but missing
values for the second followup period. The number of participants in the second group, the late
dropouts, was approximately 200 (ranging from 196 to 207) and varied slightly depending on the
examination of whichever (of the five) main dependent variable.
The analysis replicated previous regressions on the values of the five main dependent variables for
the first 12-month followup but included a dummy variable for the late dropouts and an interaction
of this dummy variable with the MHTS treatment effects. The results (in Table 8-5) indicate
significantly positive average marginal effects for the late dropout dummy for two of the five
dependent variables (ER visits and psychiatric crisis or emergency visits) but only one (positive)
significant interaction average marginal effect (for hospital stays). Results for the main MHTS
effects were very similar to those reported for “Year 1” in Table 8-3.3
Table 8-5.
Regression results for average main treatment effects with late dropout dummy
and dummy interaction with treatment*
Dependent Variable
Emergency room visits – year 1**
Outpatient psychiatric crisis
or emergency center visits –
year 1
Overnight stays – year 1
Overnight days – year 1
Other clinic or mental health
provider visits – year 1
MHTS
Est.
Effect
p-value
-0.028
0.732
Late Dropout
Est.
Effect
p-value
1.535 <0.001
Dummy Interaction
Est.
Effect
p-value
-0.193
0.392
-0.247
0.114
-0.528
0.293
0.959
0.140
-0.054
-0.847
0.213
0.057
0.046
0.078
0.618
0.946
3.579
0.624
0.002
0.678
9.017
<0.001
-6.790
0.004
2.712
0.503
*Included covariates are the five baseline use measures except as noted below.
**Covariates exclude baseline other clinic or mental health provider visits and outpatient psychiatric crisis or emergency center visits
due to convergence problems.
Discussion
The findings indicate that the treatment intervention had significant impacts in reducing inpatient
hospital use, and in reducing crisis-oriented mental health outpatient services such as ER visits for
mental health reasons and visits to other psychiatric emergency providers. In contrast, estimated
3
It should be noted that the test of bias from late dropouts does raise some concerns about endogeneity, since dropping out after four quarters of
followup may be due to unobserved severity that is correlated with the (unobservable) levels of services use in Year 2. Other strategies for
comparing late dropouts and non-dropouts, such as using service use data from Medicare and Medicaid claims for the followup period, would
provide a firmer basis for assessing late dropout bias.
Chapter 8: Utilization of Services
8-12
impacts on hospital stays for mental health problems were negative but not significant. The
significance of the negative estimated MHTS impacts on hospital inpatient days for mental health
treatment varied with the choice of included covariates; they were significant or nearly significant in
some models but not in others. None of the negative estimated impacts on ER visits for all reasons
ever approached significance.
Estimated MHTS effects on visits to other clinics or mental health providers were strongly positive.
Analyses of data on types of services provided by individual providers at these visits show positive
impacts of the MHTS on all types of services. This view is consistent with the nature of the
treatment intervention, which included vocational services using the IPS model as well as systematic
medication management and increased insurance coverage for all types of behavioral health service.
This pattern of results (i.e., negative impacts on hospital use and crisis-oriented mental health
services) provides support to the notion that the costs of implementing the MHTS intervention
were offset (at least in part) by reductions in costs for other services from non-MHTS providers. In
some cases, these cost reductions could have been substantial. For example, the average reduction in
hospital days, due to the MHTS, of approximately 0.9 days per year implies cost savings of $1,800
per year per person.4 Sustaining these annual cost savings over a longer period of time would, of
course, increase their magnitude. While the MHTS data do not allow for projecting savings over a
longer time period, it is interesting to note that the magnitude of the MHTS negative impacts on
days of inpatient hospital use appear to be roughly the same in the first and second years of the
MHTS followup period.
4
This calculation assumes a cost per day of hospital care of $2,000, which is approximately equal to the 2009 figure reported by the Agency for
Healthcare Research and Quality.
Chapter 8: Utilization of Services
8-13
Chapter 9
Study Limitations, Key Findings, and Policy Implications
The Social Security Administration (SSA) initiated the Mental Health Treatment Study (MHTS) to
assess the impact of access to evidence-based mental health treatment and supported employment
(SE) services on the employment, health, and quality of life of Social Security Disability Insurance
(SSDI) beneficiaries with schizophrenia or an affective disorder. The study design included input
from an expert panel to enhance the validity and utility of the results. Westat and its subcontractors
accepted the design, and with a few negotiated modifications, implemented the design. The earlier
chapters described the elements of the study with each contributing key pieces of information to the
overall mosaic. They provide important details needed to interpret the study results and understand
the implications for new policy initiatives.
This chapter revisits the original research questions posed by SSA, identifies study limitations,
presents a summary of the key findings, and consolidates the findings and information from the
previous chapters into a set of policy implications. No specific policies are identified nor are any
recommended for change. Instead, the policy implications section intends to give a deeper
understanding of what the study has to say about return to work of SSDI beneficiaries with
schizophrenia or an affective disorder.
Motivation for the MHTS came from a variety of sources. However, what appears to be paramount
was the availability of evidence-based treatments and employment services that held the promise of
support for the return to work efforts of SSDI beneficiaries with psychiatric impairments. In 2005,
SSA released a request for proposals (RFP) to conduct the MHTS. The RFP described the following
three research questions, which the study would answer:
1.
To what extent does delivering appropriate mental health treatment and employment
supports lead to better employment, health status, and quality of life among SSDI
beneficiaries with schizophrenia or an affective disorder?
2.
What programmatic disincentives exist that create barriers for title II beneficiaries with
schizophrenia or an affective disorder to return to work?
3.
What specific programmatic changes would support the efforts of SSDI beneficiaries
with schizophrenia or an affective disorder to sustain competitive employment?
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-1
It is important to remember these questions as they drove the development of the study design, as
well as the operational and analysis plans. These questions now serve to focus communication of the
findings and policy implications. The following sections attempt to answer these three questions,
beginning with a brief overview of the study limitations.
Study Limitations
SSA has a keen interest in external validity and the extent to which the MHTS findings generalize to
the specific population of SSDI beneficiaries with schizophrenia or an affective disorder, and to the
existing system of community mental health services in the United States. SSA is concerned with
internal validity of the study and the extent to which the findings can support inferences about the
effectiveness of the intervention, and the likelihood that changes in policy or programmatic guidance
would produce similar results. A brief discussion of each concern follows.
External Validity
There are two main topics of concern about external validity. One topic is selection of the study
sites and their representativeness of existing community mental health centers in the United States.
The other is the selection of SSDI beneficiaries and their representativeness of the SSDI rolls. This
latter question concerns the extent to which beneficiaries who participated in the MHTS represent
the universe of beneficiaries with schizophrenia or an affective disorder in the SSDI program.
Site representativeness. Community mental health centers recruited for the study were already
familiar with Individual Placement and Support (IPS), and capable of having the complex
interventions up and running within six months of the contract start date. Further, they were willing
to establish new services to meet the needs of the MHTS treatment intervention (rather than simply
redirecting services already in place). As a result, the study sites were not typical of mental health
service organizations across the country, but neither were they atypical, such as the academic
community mental health centers offering highly specialized services. None of the study sites were
affiliated with a major university hospital or clinic. At least two of the sites did not provide mental
health services directly. Instead, they negotiated provision of the services with other agencies. Even
with the intent of selecting sites that could implement the intervention quickly, as described below,
there was variability in implementation and ability to achieve fidelity in the delivery of evidence-
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-2
based services. This variability may indicate that the sites were not so carefully handpicked as to
threaten generalizability of the findings.
The study sites did not come from a randomly selected list of community mental health service
providers across all regions of the United States. Such a selection process might have increased the
external validity of the study, but the MHTS investigative team concluded (based on considerable
research evidence) that it was not feasible, given the short startup period permitted by SSA. In
addition, there was fear that this process would result in an unknown number of sites failing
completely, as not all sites would have the capability to provide access to the designated treatment
intervention services. As it was, two sites ended recruitment (in essence, failed) within the first year
of the study. The purposive site selection process meant that the study did not have as much
regional diversity as could have been achieved with stratified random site selection. In fact, finding
sites with any form of IPS SE in some states was a serious challenge. Nevertheless, there were study
sites in almost every region of the country, except for the southwest, and the beneficiary sample did
achieve a diverse set of demographic characteristics in terms of race, education, and marital status.
For example, 40 percent of the sample comprised individuals who were non-white or Hispanic.
Beneficiary representativeness. It is important to understand who was interested in participating
in the MHTS, who eventually agreed to join the study, and who remained engaged throughout.
Several selection points in the recruitment process introduced potential sources of bias and threats
to external validity. The investigators made an effort to understand these potential sources of bias
and correct for them where possible. Study site staff received training to implement extensive
standardized recruitment and enrollment procedures. The procedures described how each
beneficiary should hear about the study and how he or she should be offered the opportunity to
participate in it. Thus, to the extent possible, all beneficiaries had equal opportunity to learn about
the study and to enroll in it if desired. The standardized procedures were important for documenting
and estimating the “take-up rate” in the study. This information about MHTS participation is useful
for understanding potential enrollment in a future program.
Internal Validity
As discussed in Chapter 2, the MHTS employed random assignment of beneficiaries within sites to
the treatment or control group. This procedure alone increased the internal validity of the
observations and the ability to make inferences about the effectiveness of the treatment
intervention. One key characteristic of the study design was that the treatment intervention was
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-3
complex with many moving parts. Primarily by design, and for cost reasons, it was not possible to
determine which component(s) of the treatment intervention might predict outcomes. The MHTS
informs SSA about the effects of access to the multi-component intervention as a whole. Thus,
future replications expecting to achieve the same results will need to implement all of the
components.
The intent-to-treat (ITT) design feature focuses the interpretation of the results on the entire sample
of study participants—including all beneficiaries who had access to the treatment intervention
services without regard to the extent to which they might have participated in those specific services.
This feature of the design had potential to reduce the perceived impact of the intervention.
As discussed below, these potential limitations may bias the MHTS results in one direction or
another. The study recruited sites that were likely to be better than the average community mental
health service organization but not as good as the early adopters of IPS who participated in the first
demonstrations of the effectiveness of IPS SE or systematic medication management (SMM).
Selecting participants from the SSDI rolls and not from individuals already in mental health services
and vocational rehabilitation (who were the subjects in previous studies of IPS SE or SMM)
introduced other potential influences on the results. SSDI beneficiaries had more work experience
than many of the participants in earlier studies of IPS SE, thus it was reasonable to expect that both
the control group and the treatment group would demonstrate better employment outcomes than
those observed in earlier studies. Yet recruiting individuals with less connection to the mental health
service system might mean worse outcomes than expected from prior studies of IPS SE.
In reality, the extremely high percentages of SSDI beneficiaries in the study with serious physical
health problems and medical conditions (such as morbid obesity) made clear what may be the most
important difference between the SSDI population and populations comprising previous
demonstrations of IPS and SMM effectiveness. Due primarily to their co-occurring physical health
conditions, SSDI beneficiaries in the MHTS presented far more complex and serious disabilities to
the study sites when compared to subjects with severe mental illness in previous demonstrations, or
other clients at participating study sites. This theme became an ongoing concern throughout as the
study sites attempted to work with the SSDI population recruited and randomized to the treatment
group.
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-4
Key Findings
Findings presented below represent analyses conducted on data collected from all sources during the
study.
Enrollment and participation. The findings from this chapter focus on the extent of participation
in the MHTS from the SSDI rolls, and reasons why beneficiaries chose not to enroll in the study.
The key findings are presented below.
1.
Based on the sample of potential enrollees, the findings suggest that SSA could expect 14
percent of the SSDI beneficiaries with schizophrenia or an affective disorder might
enroll in an MHTS-like program. That percentage could jump to 26 percent if SSA were
to target a specific subpopulation of beneficiaries.
2.
The strongest predictors of enrollment were those related to prior work activity
reported in SSA’s administrative records, including beneficiaries who report earnings
within the past 6 months (prior to enrollment), had assigned their Ticket (in the Ticket
to Work program), or had a Trial Work End Date within the past 3 years.
3.
Beneficiaries enrolling in the study were younger, had been on the SSDI rolls a shorter
period, were only on SSDI (as opposed to concurrent SSI beneficiaries), and did not
have a RepPayee.
4.
The most commonly recorded single reason for beneficiaries not enrolling in the study
was general disinterest (37%). However, nearly 40 percent of all beneficiaries with a
recorded reason, indicated work-related or physical health issues were reasons they did
not enroll in the study.
5.
The randomization procedure was successful, as no statistical differences existed
between the treatment and control groups at baseline on key outcome measures,
including recent employment (in the 2 years prior to enrollment), physical health status,
and mental health status.
Outcomes. The following findings represent analyses of interview data on the primary outcomes of
interest in the study, including employment rate and earnings; mental health status; physical health
status; and quality of life. For purposes of brevity, no secondary outcome analyses appear here.
6.
The 24-month employment rate for the treatment group was 61 percent compared to
40 percent for the control group. The difference between these percentages was
significant (p-value < 0.001).
7.
Analyses of average earnings (in the past month averaged over eight interviews) show
that treatment group participants earned significantly more than did control group
participants. One analysis concerns comparisons between the unconditional means of
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-5
the treatment and control groups ($148 vs. $97, p-value < 0.001). A second analysis
concerns the comparison between conditional means (i.e., including only those
participants with non-zero earnings) of the treatment and control groups ($251 vs. $228,
p-value < 0.001). A third analysis concerns the percentages of participants in the
treatment and control groups with earnings (59% vs. 43%, p-value < 0.001).
8.
Eight percent of the study participants (both treatment and control) showed average
earnings over the 24-month study period that exceeded the current level of substantial
gainful activity (SGA). Participants in the treatment group did not experience an
increase in work that SSA considers SGA when compared to participants in the control
group. Neither did participants in the treatment group experience a reduction in benefit
payments when compared to participants in the control group.
9.
Measured at baseline and again at study exit, the treatment group showed a significant
improvement over the control group in mental health status (p-value < 0.001).
10.
Measured at baseline and again at study exit, both the treatment and control groups
showed a slight decline in physical health status, however, the changes between the
groups were not statistically significant (p-value = 0.924).
11.
Measured at baseline and again at study exit, the treatment group showed a significant
improvement over the control group in quality of life (p-value < 0.001).
Implementation of SE. The key findings are:
12.
Eighty percent or more of the study sites achieved a high level of IPS program
implementation (i.e., met the documented standard for high fidelity). This high level of
implementation persisted across the entire study period.
13.
The level of unengagement in employment services among treatment group participants
was relatively low overall (~10%).
Implementation of SMM. The key findings are:
14.
Concordance between the SSA diagnostic category and the Structured Clinical Interview
for DSM-IV Axis I Disorders (SCID) diagnosis was greater than 80 percent.
15.
More than 87 percent of treatment group participants had at least one physical health
condition, and 69 percent had two or more. More than half of all participants in the
treatment group had a body mass index (BMI) in the obese range.
16.
The quality of SMM varied considerably across study sites. The off-site location of
many prescribers affected this measure. For purposes of conducting the MHTS, the
decision to allow participants to remain with their outside prescribers was reasonable.
However, the goal of MHTS was to deliver an integrated package of services to
participants. Having off-site prescribers presented great difficulties in integrating the
SMM components with other treatment components.
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-6
Health care and SE financing. The Health Benefits Plan (HBP) paid for behavioral health and SE
services, out-of-pocket expenses, and health insurance premiums. The goal was to leave the
treatment group participant with no costs associated with services provided with the intervention
package. The key findings are:
17.
Spending through the HBP averaged $6,986 per study participant per year for those
treatment group participants who completed the study. Spending was less ($6,342) for
the overall treatment group population (which includes those participating for less than
24 months). These figures do not include cost of services related to the Nurse Care
Coordinator (NCC) role.
18.
The overall spending distribution included over 70 percent for SE services; 11 percent
for health insurance premiums; 8 percent for medication prescriptions; 7 percent for
behavioral health services; and less than 3 percent for essential work-related expenses,
transportation, and other miscellaneous expenses. SE services were the only services
completely paid by the study.
Utilization of services. Self-report interviews also asked about hospital admissions, emergency
room (ER) visits, outpatient psychiatric emergency or crisis visits, and regular provider visits. The
key findings associated with service use include the following:
19.
The treatment intervention had significant positive impacts in reducing inpatient
hospital use (for both admissions and number of days) and outpatient psychiatric
emergency or crisis visits.
20.
As expected with the intervention package, treatment group participants showed
significantly higher use of regularly scheduled clinic or mental health visits.
21.
The average reduction in hospital days was 0.9 days per year, which translated into
approximately $1,800 per year per person. Sustaining these costs over longer periods
would increase their magnitude.
These findings answer SSA’s first question whether the treatment intervention leads to better
employment, health, and quality of life. The answer is, it did. The findings presented above are clear
and unambiguous about the impact of the treatment intervention on SSDI beneficiaries with
schizophrenia or an affective disorder who express an interest in working. The treatment group
attained significantly better outcomes compared to the control group, and they did it repeatedly with
nearly every outcome variable. With no demonstrated differences at baseline between the treatment
and control groups on key outcome measures, at study exit the treatment group demonstrated
statistically significant improvements or differences over the control group. Rarely did the control
group outperform the treatment group, and when it did, it was an isolated finding. In Chapter 4,
these differences between the treatment and control groups also held in many, if not most,
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-7
secondary analyses of the employment-related outcomes (such as total months employed, average
hours per week at main job, etc.), as well as the analyses of relevant subgroups (such as diagnosis,
gender, age, and education).
The findings also point to the extent to which the treatment intervention would affect the target
SSDI population. The enrollment and participation findings make clear that a significant segment of
the SSDI population with schizophrenia or an affective disorder would enroll in an MHTS-like
program.
These results offer SSA a significant opportunity to develop policies to assist these beneficiaries to
return to work. What effects or influences these findings may have on the SSDI rolls, the Trust
Fund, or cost offsets for providing MHTS-like services are unclear and outside the purview of this
report. However, it is abundantly clear that significant improvement in employment, mental health,
and quality of life is possible with improvements in access to the package of services delivered
through the treatment intervention.
Policy Implications
The findings from the study have implications for policies related to the SSDI program, as well as
for employment policies of other federal agencies as these pertain to individuals with severe mental
illness. It is clear from the research that SSA cannot solve all of the problems of providing access to
all of the services tested in the MHTS with changes to its SSDI program policies only. The services
implemented in the study are relevant and should be of interest to many federal agencies that serve
individuals with serious mental illness, including schizophrenia or an affective disorder.
The purpose of the following comments is to inform policy discussions at SSA and other relevant
agencies. The intent of the discussion is to provide answers to the remaining SSA research questions
about programmatic disincentives and programmatic changes needed to support SSDI beneficiaries
return to work. The remaining sections offer a question-and-answer format to make the information
more accessible.
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-8
Is there significant interest among SSDI beneficiaries with schizophrenia or an affective
disorder to access an MHTS-like program?
The study findings presented above, and further detailed in Chapter 3, make it clear that a
substantial percentage of beneficiaries with schizophrenia or an affective disorder would consider
participating in an MHTS-like program. The numbers may approach a quarter or more of the
targeted beneficiaries on the SSDI rolls. Today there are nearly 2.25 million SSDI beneficiaries with
a psychiatric disorder. The largest portion of them includes beneficiaries with schizophrenia or an
affective disorder. The MHTS finding that 14 percent of beneficiaries could be expected to take
advantage of MHTS-like services (Chapter 3) suggests that perhaps 306,000 SSDI beneficiaries could
be affected. Taking the finding further, the data suggest that more than 60 percent of those who go
on to participate in such services would find a job within a 24-month period. While this is
hypothetical, the study findings point out that there is a significant number of SSDI beneficiaries
who want to work and can work if given access to an MHTS-like intervention package.
What essential services or features of services are required to achieve MHTS-like results?
The intervention package in the MHTS was rich in services and benefits. The following list and
descriptions provide details of the services and benefits, as well as the features considered essential
to achieve the results attained in this study:
Essential services. The primary services included evidence-based SE (the IPS model), SMM, other
behavioral health (OBH) services (such as therapy, counseling, substance abuse counseling, etc.),
benefits counseling, and modest monetary support to pay for inaccessible services.
Community mental health centers. The usual mental health care setting that can offer both the
comprehensive range of services needed to treat mental illness and provide integrated vocational
services is the community mental health center. The experience of the MHTS suggests that the
community mental health center is ideal for delivery of these essential services. As noted above,
community mental health centers hosted the study in every site with two exceptions. One of the two
exceptions was a vocational services program with a negotiated contract for mental health services
with the county mental health board. The other site was a housing program for the homeless and
mentally ill. This program offered employment services and negotiated mental health services in the
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-9
community. All of the community mental health centers offered a wide range of mental health
services, case management, and employment services. With only one exception, all of the
community mental health centers had experience providing IPS SE services. However, not all of
them had the documentation to support that they were performing at high fidelity at the time the
study began.
Need for a care coordinator. Use of a care coordinator to facilitate SMM was a key feature of the
mental health services provided in the study. The SMM program in the MHTS used the NCC to
facilitate and promote prescriber use of evidence-based guidelines and recommendations for
medication management of severe and persistent mental illnesses. As noted in Chapter 6, the central
goal of SMM was to provide prescribers with pertinent, updated information about the patient and
medication options at the time of each medication visit with the prescriber. In the ideal SMM model,
the NCC and prescriber function in tandem with the NCC first assessing the patient’s status with
regard to medication use, medication effects, and medication-related symptom outcomes. The
prescriber received this information, along with any recommendations for change based on
medication management guidelines and expert consultation. The prescriber evaluated the
information and the patient, and made final judgments about medication management. The exact
division of labor between NCC and prescriber was determined at the local level. In clinic and office
settings, implementation of SMM requires systemic changes in personnel roles and in the recording
of information obtained from patients. The study findings note that 90.4 percent of participants
were engaged in SMM at some point in the study. Further, the study found that 53 percent of on-site
prescribers engaged the NCC compared to only 11 percent of off-site prescribers. This latter finding
suggests that on-site prescribers, while not a requirement, are more likely to be fully engaged in the
SMM process.
Out-of-pocket expenses. Payment for out-of-pocket mental health and essential work-related
expenses is an important feature that improved access to services and jobs for study participants.
Unfortunately, it is not a simple decision for some SSDI beneficiaries (as described below in the
section on disincentives to return to work). The decision to make a co-payment is often complex
and difficult for these beneficiaries. In the MHTS, payments for out-of-pocket expenses for
participants who were trying to return to work created opportunities that, otherwise, would not have
existed.
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-10
How much did the study spend on each treatment group participant to achieve the study
results?
Understanding expenditures associated with the intervention helps to put into perspective the
impact of the treatment intervention on the study outcomes. The HBP provided all payments for
participants in the study. Eight principles guided the development of health care and SE financing
for individual participants and related payment decisions. These principles included the following:
1.
Provide all participants a comprehensive health care coverage package.
2.
Ensure that participants receive needed care.
3.
Ensure that participants receive evidence-based treatments.
4.
Cover SE in full.
5.
Cover all other approved MHTS behavioral health care expenditures in full.
6.
Utilize scarce resources wisely.
7.
Minimize participant up-front out-of-pocket expenditures.
8.
Collect research data on encounters and expenditures.
Overall, participants spent an average of $6,986 per year. Of course, these expenditures do not
reflect all of the participant costs for services needed during participation in the study. NCC
services, for example, were paid through the study, and not as part of the HBP. Medicare or
Medicaid paid for many of the mental health and general health services required by participants.
However, it is not clear how important those costs are to a discussion of how to provide MHTS-like
services to the broader population of participants with psychiatric impairments reflected on the rolls.
That is, except for participants in the 24-month waiting period for Medicare.
It is noteworthy that fewer hospital stays and fewer psychiatric emergency or crisis visits may offset
some of the HBP expenditures. Further exploration of the MHTS data may clarify the potential for
health care.
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-11
What programmatic disincentives exist that create barriers for title II participants with
schizophrenia or an affective disorder to return to work?
The design of the MHTS did not permit any direct tests of potential programmatic barriers to work.
Given the many components of the intervention package, it was not possible to parse them into
discrete testable components. Each major element of the intervention package was a solution to a
known barrier to return to work. The design contract for the MHTS (Aron, Burt, & Wittenburg,
2005) specifically built the study to test the impact of removing these barriers—as a whole. Thus, it
was not possible to isolate the individual contributions of any particular feature of the intervention
to the study results, including access to comprehensive health insurance support, access to SE
services, SMM, or waiver of the medical Continuing Disability Review.
Three general barriers describe the kinds of problems SSDI beneficiaries faced in gaining access to
needed health care and employment programs and supports. They include access to health care,
access to evidence-based employment services, and complex co-morbid physical conditions. They
are not disincentives in the sense of the well-known disincentives associated with fear of loss of
benefits—or more likely fear of not being able to get back on benefits once they are lost. Instead,
they represent barriers, or hurdles, that participants cannot negotiate alone without some sort of
programmatic intervention. Below we provide a discussion of each set of barriers as we have come
to understand them.
Under current conditions, SSDI beneficiaries have insufficient access to health care
programs, services, and treatments. The study revealed many instances where insufficient access
to programs, services, and treatments was a problem. Expenditures through the HBP made it clear
that obtaining payment for all behavioral health care costs, along with payment of insurance
premiums, removed some of the putative barriers to participation in active efforts to return to work
for individuals in the treatment group. These payments represented less than 10 percent of the total
HBP expenditures, suggesting that across all participants in the treatment group, gaining access to
needed health care amounted to approximately $53 per month. A remarkably low cost for improving
access to needed health care.
The vast majority of participants in the treatment group at the time of study entry had
comprehensive health care coverage through Medicare or Medicaid, while a small fraction did not.
Approximately 7 percent (74) participants required enrollment in one or more Parts (A, B, or D) of
Medicare (See Chapter 7).
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-12
Reports from participants in the treatment group suggested that in the past, the cost of insurance
co-pays (for health care visits, prescription medications, etc.) kept some participants from seeking
treatment. In general, it is clear that such decisions to seek treatment were complicated and highly
individual. For example, a participant related a story that prior to the study she chose not to refill a
prescription for her psychiatric medication because it was too expensive given her financial
condition. However, upon further discussion, it became clear that the decision was more
complicated than that. She felt the medication was not effective (had unpleasant side effects and was
not sufficiently reducing symptoms) and, therefore, was not worth the cost of the co-pay to refill it.
Many previous trips to the doctor to find a better medication proved expensive and ineffective. It
was no longer worth the effort required to seek a solution. It was cheaper and easier to do nothing.
In sum, the participant decided that refilling the prescription was not worth the investment of scarce
dollars. This example reveals the importance and roles of various MHTS elements, such as covering
co-pays, assisting with travel to treatment visits when indicated, and SMM in the overall scheme of
return to work. No single MHTS benefit could have solved the problem presented by this
participant.
One ongoing concern throughout the study was the role that high-cost psychiatric medications
played in finding effective treatments for participants in the treatment group. Several situations were
problematic. One clear problem was the effect of the so-called “donut hole” on medication use.
NCCs reported that prior to the MHTS when beneficiaries came up against the “donut hole,” they
would stop refilling their prescription(s) for psychiatric (and other) medication(s) because the high
costs did not fit within their budget. Another problem reported by some beneficiaries was that prior
to the study, prescribers were reluctant to prescribe some (potentially more effective) medications
due to the high cost to the patient. A third problem was finding a Part D plan that covered the
preferred medications. During the study, this always presented a challenge, and required more
efforts from the Westat Insurance Planner (WIP) than any other insurance issue. A change in the
Part D prescription medication plan required waiting until the open enrollment period. This
situation limited prescribers in their options to find an optimal treatment for their patient.
Fortunately, the MHTS paid for a prescription medication whenever a participant’s plan did not
cover it. Then, during the open enrollment period, the WIP would work with the NCC and the
participant to find a more acceptable plan. The findings would not likely be the same without the
support of the study to pay for medications when they were not covered, or high co-pays.
The study paid special attention to receipt of mental health case management services, because case
management is a key for achieving adequate mental health treatment (Rapp & Goscha, 2004). The
fact that only 54 percent of participants received mental health case management, and of these, 28
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-13
percent received their case management services off-site, is far below the expected rate in high
fidelity IPS programs serving clients with severe mental illness. Several factors may account for this
finding. First, some participants did not need case management services. Unfortunately, study
attempts to measure need were not successful, so there was no way of reliably estimating need.
Second, some participants received case management assistance from the NCC. The study did not
include a direct measure of NCC case management activity, so no data exists on the extent of NCC
case management. Nevertheless, given the competing responsibilities, NCCs must have played a
limited role in case management—both in number of participants assisted and in the range of case
management activities performed. The Quality Management Program Director site reports reinforce
the conclusion that at least at some study sites, case management services were insufficient. Since
case management is a vital part of an integrated IPS team and was often unavailable, it may be a
missing link in the engagement process of participants receiving IPS services. This may have been
more of a factor in this study than in past IPS interventions, because many of the participants were
not engaged in community mental health services when they entered the study. In usual IPS practice,
clients are typically first engaged in mental health treatment and then referred to IPS. As has been
noted many times, randomization into the treatment intervention occurred without participants’ first
receiving mental health treatment at the study site. Some were not receiving any treatment, and
many were being treated elsewhere.
Each of the problems presented above create distinct barriers to effective treatment, better
functioning and, ultimately, to employment. No one barrier by itself is particularly troublesome.
However, as a group, they represent formidable obstacles to return to work.
Under current conditions, SSDI beneficiaries lack access to evidence-based SE services in
community mental health centers. Analysis of the health services utilization data reveals that the
treatment group used more vocational services than the control group. This was expected, given the
intervention package included the provision of evidence-based IPS SE. These services, which
emphasized competitive employment, clearly contributed to the positive employment results
attained in the study. However, as discussed elsewhere in this report, community mental health
centers do not generally provide SE services. Payment for these services is limited to Medicaid. In
fact, even Medicaid does not cover these services in every state. A recent report to the Assistant
Secretary for Planning and Evaluation by Karakus, Frey, Goldman, Fields, & Drake (2011) describes
both the difficulty of funding these services in general, and more specifically, the need for a federal
financing strategy to provide them in community mental health centers. The report points out that
Medicaid funding, as designed, is inadequate for this purpose. The result is most community mental
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-14
health centers cannot afford to employ the staff (employment specialists, supervisors, etc.) necessary
to provide SE services.
Many SSDI beneficiaries with schizophrenia or an affective disorder have complex comorbid physical conditions that impede efforts to return to work. Physical impairments created
by a wide variety of health conditions were serious deterrents to employment efforts made by study
participants, NCCs, and employment specialists. As reported above, 87 percent of the participants in
the treatment group had at least one co-morbid physical health condition, 69 percent had two or
more, and more than half were obese. Frequent comments on conference calls throughout the study
noted that the SSDI beneficiary population in the study was much less healthy and had more
problematic health conditions than did the populations normally served by the mental health
centers. These comments corroborated the data collected from treatment group participants and
formal documentation of why some beneficiaries do not return to work (discussed in Chapters 4, 5,
and 6). In fact, good physical health (as measured by the SF-12) frequently came up as a predictor of
employment (obtained employment, steady worker, or number of months employed), suggesting
that there was enough variability in physical health to pick it up in multivariate analyses.
The implications of this fact are twofold. First, this situation reinforces the need for improved health
care access. As noted above, these beneficiaries sometimes weigh the cost of seeking health care
with the expectation of a good or poor outcome. As already mentioned, sometimes the decision is
made to forego refilling a prescription or doctor visit, when it is not likely in the best interest of
physical health. Second, working on physical health, mental health, and employment all at once likely
slows the return to work process. This was the case for a number of beneficiaries in the study.
Chapter 9: Study Limitations, Key Findings, and
Policy Implications
9-15
References
Aron, L., Burt, M., & Wittenburg, D. (2005). Recommendations to the Social Security Administration on the
design of the Mental Health Treatment Study (MHTS). Washington, DC: Urban Institute.
Becker, D.R., Bebout, R.R., & Drake, R.E. (1998). Job preferences of people with severe mental
illness: A replication. Psychiatric Rehabilitation Journal, 22(1), 46-50.
Becker, D.R., Bond, G.R., McCarthy, D., Thompson, D., Xie, H., McHugo, G.J., & Drake, R.E.
(2001). Converting day treatment centers to supported employment programs in Rhode
Island. Psychiatric Services, 52(3), 351-357.
Becker, D.R. & Drake, R.E. (2003). A working life for people with severe mental illness. New York: Oxford
University Press.
Becker, D.R., Drake, R.E., Bond, G.R., Nawaz, S., Haslett, W.R., & Martinez, R.A. (2011). A mental
health learning collaborative on supported employment. Psychiatric Services, 62, 704-706.
Becker, D.R., Drake, R.E., Farabaugh, A., & Bond, G.R. (1996). Job preferences of clients with
severe psychiatric disorders participating in supported employment programs. Psychiatric
Services, 47, 1223-1226. 22(1), 24-29.
Becker, D.R., Swanson, S., Bond, G.R., & Merrens, M.R. (2008). Evidence-based supported employment
fidelity review manual. Lebanon, NH: Dartmouth Psychiatric Research Center.
Becker, D.R., Whitley, R., Bailey, E.L., & Drake, R.E. (2007). Long-term employment outcomes of
supported employment for people with severe mental illness. Psychiatric Services, 58, 922-928.
Becker, D.R., Xie, H., McHugo, G.J., Halliday, J., & Martinez, R.A. (2006). What predicts supported
employment program outcomes? Community Mental Health Journal, 42, 303-313.
Bedell, J.R., Draving, D., Parrish, A., Gervey, R., & Guastadisegni, P. (1998). A description and
comparison of experiences of people with mental disorders in supported employment and
paid prevocational training. Psychiatric Rehabilitation Journal, 21(3), 279-283.
Berkowitz, M. (2003). The Ticket to Work Program: The complicated evolution of a simple idea. In
K. Rupp & S.H. Bell (Eds). Vocational rehabilitation: Will provider incentives work for Ticket to
Work. Washington, DC: Urban Institute.
Bond, G.R. (2007). Modest implementation efforts, modest fidelity, and modest outcomes. Psychiatric
Services, 58, 334.
Bond, G.R. (2004). Supported employment: Evidence for an evidence-based practice. Psychiatric
Rehabilitation Journal, 27, 345-359.
Bond, G.R. (1998). Principles of the Individual Placement and Support model: Empirical support.
Psychiatric Rehabilitation Journal, 22(1), 11-23.
References-1
Bond, G.R., Becker, D.R., & Drake, R.E. (2011). Measurement of fidelity of implementation of
evidence-based practices: Case example of the IPS Fidelity Scale. Clinical Psychology: Science and
Practice, 18, 126-141.
Bond, G.R., Becker, D.R., Drake, R.E., & Vogler, K.M. (1997). A fidelity scale for the Individual
Placement and Support model of supported employment. Rehabilitation Counseling Bulletin, 40,
265-284.
Bond, G.R., & Drake, R.E. (2008). Predictors of competitive employment among patients with
schizophrenia. Current Opinion in Psychiatry, 21, 362-369.
Bond, G.R., Drake, R.E., & Becker, D.R. (2008). An update on randomized controlled trials of
evidence-based supported employment. Psychiatric Rehabilitation Journal, 31, 280-290.
Bond, G.R., Drake, R.E., McHugo, G.J., Rapp, C.A., & Whitley, R. (2009). Strategies for improving
fidelity in the National Evidence-Based Practices Project. Research on Social Work Practice, 19,
569-581.
Bond, G.R., Drake, R.E., Mueser, K.T., & Becker, D.R. (1997). An update on supported
employment for people with severe mental illness. Psychiatric Services, 48, 335-346.
Bond, G.R., Evans, L., Salyers, M.P., Williams, J., & Kim, H.K. (2000). Measurement of fidelity in
psychiatric rehabilitation. Mental Health Services Research, 2, 75-87.
Bond, G.R., Xie, H., & Drake, R.E. (2007). Can SSDI and SSI beneficiaries with mental illness
benefit from evidence-based supported employment? Psychiatric Services, 58, 1412-1420.
Brekke, J.S. (1988). What do we really know about community support programs? Strategies for
better monitoring. Hospital and Community Psychiatry, 39, 946-952.
Burns, T., Catty, J., White, S., Becker, T., Koletsi, M., Fioritti, A., Rössler, W., Tomov, T., van
Busschbach, J., Wiersma, D., & Lauber, C. (2009). The impact of supported employment
and working on clinical and social functioning: Results of an international study of Individual
Placement and Support. Schizophrenia Bulletin, 35, 949-958.
Campbell, K., Bond, G.R., Drake, R.E., McHugo, G.J., & Xie, H. (2010). Client predictors of
employment outcomes in high-fidelity supported employment: A regression analysis. Journal
of Nervous and Mental Disease, 198, 556-563.
Campbell, K., Bond, G.R., Gervey, R., Pascaris, A., Tice, S., & Revell, G. (2007). Does type of
provider organization affect fidelity to evidence-based supported employment? Journal of
Vocational Rehabilitation, 27, 3-11.
Carpenter, V., & Perkins, D.V. (1997, November 8). Supported employment as a transition from state
hospital to community. Paper presented at the American Public Health Association,
Indianapolis, IN.
Conover, W.J., & Iman, R.L. (1981). Rank transformation as a bridge between parametric and
nonparametric statistics. The American Statistician, 35, 125-129.
References-2
Cook, J.A., Lehman, A.F., Drake, R., McFarlane, W.R., Gold, P.B., Leff, H. S., Blyler, C., Toprac,
M.G., Razzano, L.A., Burke-Miller, J.K., Blankertz, L., Shafer, M., Pickett-Schenk, S.A., &
Grey, D.D. (2005). Integration of psychiatric and vocational services: A multisite
randomized, controlled trial of supported employment. American Journal of Psychiatry, 162,
1948-1956.
Corrigan, P.W., Mueser, K.T., Bond, G.R., Drake, R.E., & Solomon, P. (2008). Principles and practice of
psychiatric rehabilitation: An empirical approach. New York: Guilford Press.
Cradock, J., Young, A.S., et al. (2001). The accuracy of medical record documentation in
schizophrenia. Journal of Behavioral Health Services and Research, 28, 456-465.
Crane-Ross, D., Roth, D., & Lauber, B.G. (2000). Consumers' and case managers' perceptions of
mental health and community support service needs. Community Mental Health Journal, 36,
161-178.
Danziger, S., Frank, R., & Meara, E. (2009). Mental illness, work and income support programs.
American Journal of Psychiatry, 166(4), 398-404.
Dickinson, D., Ramsey, M., & Gold, J. (2007). Overlooking the obvious: A meta-analytic
comparison of digit symbol coding tasks and other cognitive measures in schizophrenia.
Archives of General Psychiatry, 64, 532-542.
Drake, R.E., Becker, D.R., Bond, G.R., & Mueser, K.T. (2003). A process analysis of integrated and
non-integrated approaches to supported employment. Journal of Vocational Rehabilitation, 18,
51-58.
Drake, R.E., Becker, D.R., Goldman, H.H., & Martinez, R.A. (2006). Best practices: The Johnson &
Johnson—Dartmouth Community Mental Health Program: Disseminating evidence-based
practice. Psychiatric Services, 57, 302-304.
Drake, R.E., & Bond, G.R. (2010). Implementing integrated mental health and substance abuse
services. Journal of Dual Diagnosis, 6, 251-262.
Drake, R.E., & Essock, S.M. (2009). The science-to-service gap in real-world schizophrenia
treatment: The 95% problem. Schizophrenia Bulletin, 35, 677-678.
Drake, R.E., Essock, S.M., & Bond, G.R. (2009). Implementing evidence-based practices for people
with schizophrenia. Schizophrenia Bulletin, 35, 704-713.
Drake, R.E., McHugo, G.J., Bebout, R.R., Becker, D.R., Harris, M., Bond, G.R., & Quimby, E.
(1999). A randomized clinical trial of supported employment for inner-city patients with
severe mental illness. Archives of General Psychiatry, 56, 627-633.
Fabian, E.S. (1989). Supported employment for severely psychiatrically disabled persons: A
descriptive study. Psychosocial Rehabilitation Journal, 13(2), 53-60.
References-3
Frounfelker, R.L., Wilkniss, S., Bond, G.R., Drake, R.E., & Devitt, T.S. (in press). Interest, access,
and outcomes of supported employment services for clients with co-occurring disorders.
Psychiatric Services.
Gervey, R., & Kowal, H. (1994, May). A description of a model for placing youth and young adults with
psychiatric disabilities in competitive employment. Paper presented at the International Association
of Psychosocial Rehabilitation Services Conference, Albuquerque, NM.
Gold, P.B., Meisler, N., Santos, A.B., Carnemolla, M.A., Williams, O.H., & Kelleher, J. (2006).
Randomized trial of supported employment integrated with assertive community treatment
for rural adults with severe mental illness. Schizophrenia Bulletin, 32, 378-395.
Grant, B.F., Stinson, F.S., Dawson, D.A, Chou, S.P., Dufour, M.C., Compton, W., et al. (2004).
Prevalence and co-occurrence of substance use disorders and iindependent mood and
anxiety disorders: Results from the National Epidemiologic Survey on Alcohol and Related
Conditions. Archives of General Psychiatry, 61, 807-816.
Grob, G.N., & Goldman, H.H. (2006). The dilemma of federal mental health policy: Radical reform or
incremental change? New Brunswick, NJ: Rutgers University Press.
Hales, R.E, Yudofsky, S.C., Gabbard, G.O. (Eds.). (2010). Essentials of psychiatry (3rd ed). Arlington,
VA: American Psychiatric Publishing, Inc.
Hennessey, J.C., & Dykacz, J.M. (1989). Projected outcomes and length of time in disability
insurance program. Social Security Bulletin, 52(9), 2-41.
Heslin, M., Howard, L., Leese, M., McCrone, P., Rice, C., Jarrett , M., Spokes, T., Huxley, P., &
Thornicroft, G. (2011). Randomized controlled trial of supported employment in England: 2
year follow-up of the Supported Work and Needs Study. World Psychiatry, 10, 132-137.
Hollis S. & Campbell F. (1999). What is meant by intention to treat analysis? Survey of published
randomised controlled trials. British Medical Journal, 319, 670-674.
Huff, S.W., Rapp, C.A., & Campbell, S.R. (2008). Every day is not always Jell-O: A qualitative study
of factors affecting job tenure. Psychiatric Rehabilitation Journal, 31, 211-218.
Karakus M., Frey W., Goldman H., Fields S., & Drake R.E. (2010). Federal financing of supported
employment & customized employment for people with mental illness. (Prepared under contract to the
Office of the Assistant Secretary for Planning and Evaluation, U.S. Department of Health
and Human Services). Rockville, MD: Westat.
Katon, W., Von Korff, M., Lin, E., Walker, E., Simon, G.E., Bush, T., et al. (1995). Collaborative
management to achieve treatment guidelines: Impact on depression in primary care. Journal of
the American Medical Association, 273, 1026-1031.
Kessler, R.C. (2004). The epidemiology of dual diagnosis. Biological Psychiatry, 56, 730-737.
References-4
Krenzke, T., & Judkins, D. (2008). Filling in blanks: Some guesses are better than others –
Illustrating the impact of covariate selection when imputing complex survey items. Chance,
21(3), 7-13.
Kreyenbuhl, J., Buchanan, R.W., Dickerson, F.B., & Dixon, L.B. (2010). The Schizophrenia Patient
Outcomes Research Team (PORT): Updated treatment recommendations 2009. Schizophrenia
Bulletin, 36, 94-103.
Kreyenbuhl, J., Nosse, I.R., & Dixon, L.B. (2009). Disengagement from mental health treatment
among individuals with schizophrenia and strategies for facilitating connections to care: A
review of the literature. Schizophrenia Bulletin, 35, 696-703.
Lachin, J. (2000). Biostatistical methods: The assessment of relative risks (Wiley Series in Probability and
Statistics). New York: Wiley.
Latimer, E., Lecomte, T., Becker, D., Drake, R., Duclos, I., Piat, M., Lahaie, N., St. Pierre, M.S.,
Therrien, C., & Xie, H. (2006). Generalisability of the individual placement and support
model of supported employment: Results of a Canadian randomised controlled trial. British
Journal of Psychiatry, 189, 65-73.
Leff, H.S., Cook, J.A., Gold, P.B., Toprac, M., Blyler, C., Goldberg, R.W., McFarlane, W., Shafer,
M., Allen, I.E., Camacho-Gonsalves, T., & Raab, B. (2005). Effects of job development and
job support on competitive employment of persons with severe mental illness. Psychiatric
Services, 56, 1237-1244.
Lehman, A.F. (1988). A Quality of Life Interview for the chronically mentally ill. Evaluation and
Program Planning, 11, 51-62.
Lehman, A.F., Goldberg, R.W., Dixon, L.B., McNary, S., Postrado, L., Hackman, A., & McDonnell,
K. (2002). Improving employment outcomes for persons with severe mental illness. Archives
of General Psychiatry, 59, 165-172.
Lehman, A.F., Steinwachs, D.M., & the co-investigators of the PORT. (1998). Project patterns of
usual care for schizophrenia: Initial results from the Schizophrenia Patient Outcomes
Research Team client survey. Schizophrenia Bulletin, 24, 11-19.
Livermore, G.A., Goodman, N., & Wright, D. (2007). Social Security disability beneficiaries:
Characteristics, work activity, and use of services. Journal of Vocational Rehabilitation, 27, 85-93.
Loveland, D., Driscoll, H., & Boyle, M. (2007). Enhancing supported employment services for
individuals with a serious mental illness: A review of the literature. Journal of Vocational
Rehabilitation, 27, 177-189.
MacDonald-Wilson, K.L., Revell, W.G., Nguyen, N.H., & Peterson, M.E. (1991). Supported
employment outcomes for people with psychiatric disability: A comparative analysis. Journal
of Vocational Rehabilitation, 1, 30-44.
References-5
MacDonald-Wilson, K.L., Rogers, E.S., Ellison, M.L., & Lyass, A. (2003). A study of the Social
Security work incentives and their relation to perceived barriers to work among persons with
psychiatric disability. Rehabilitation Psychology, 48, 301-309.
Marker, D.A., Judkins, D.R., & Winglee, M. (2001). Imputation in large-scale complex surveys. In B.
Groves et al. (Eds.), Survey nonresponse (pp. 329-341). New York, NY: John Wiley.
McAlpine, D.D., & Warner, L. (2000). Barriers to employment among persons with mental illness: A review of
the literature. New Brunswick, NJ: Center for Research on the Organization and Financing of
Care for the Severely Mentally Ill Institute for Health, Health Care Policy and Aging
Research, Rutgers, the State University.
McHugo, G.J., Drake, R.E., Whitley, R., Bond, G.R., Campbell, K., Rapp, C.A., Goldman, H.H.,
Lutz, W.J., & Finnerty, M.T. (2007). Fidelity outcomes in the National Implementing
Evidence-Based Practices Project. Psychiatric Services, 58, 1279-1284.
McQuilken, M., Zahniser, J.H., Novak, J., Starks, R.D., Olmos, A., & Bond, G.R. (2003). The Work
Project Survey: Consumer perspectives on work. Journal of Vocational Rehabilitation, 18, 59-68.
Mechanic, D., Bilder, S., & McAlpine, D.D. (2002). Employing persons with serious mental illness.
Health Affairs, 21(5), 242-253.
Menear, M., Reinharz, D., Corbière, M., Houle, N., Lanctôt, N., Goering, P., Goldner, E.M., Kirsh,
B., & Lecomte, T. (2011). Organizational analysis of Canadian supported employment
programs for people with psychiatric disabilities. Social Science and Medicine, 72, 1028-1035.
Michie, S., Fixsen, D., Grimshaw, J.M., & Eccles, M.P. (2009). Specifying and reporting complex
behaviour change interventions: the need for a scientific method. Implementation Science, 4(40),
doi:10.1186/1748-5908-1184-1140.
Miller, A.L., Crismon, M.L, Rush, A.J. Chiles, J., Kashner, T.M., Toprac, M. et al. (2004). The Texas
Medication Algorithm Project: Clinical results for schizophrenia. Schizophrenia Bulletin, 30(3),
627-647.
Mowbray, C.T., McCrohan, N.M., & Bybee, D. (1995). Integrating vocational services into case
management: Implementation analysis of Project WINS. Journal of Vocational Rehabilitation, 5,
89-102.
Mueser, K.T., Becker, D.R., & Wolfe, R.S. (2001). Supported employment, job preferences, job
tenure and satisfaction. Journal of Mental Health, 10, 411-417.
Mueser, K.T., Becker, D.R., Torrey, W.C., Xie, H., Bond, G.R., Drake, R.E., & Dain, B.J. (1997).
Work and nonvocational domains of functioning in persons with severe mental illness: A
longitudinal analysis. Journal of Nervous and Mental Disease, 185, 419-426.
Mueser, K.T., Clark, R.E., Haines, M., Drake, R.E., McHugo, G.J., Bond, G.R., Essock, S.M.,
Becker, D.R., Wolfe, R., & Swain, K. (2004). The Hartford study of supported employment
for persons with severe mental illness. Journal of Consulting and Clinical Psychology, 72, 479-490.
References-6
Mueser, K.T., Salyers, M.P., & Mueser, P.R. (2001). A prospective analysis of work in schizophrenia.
Schizophrenia Bulletin, 27, 281-296.
Newcomb, C., Payne, S., & Waid, M.D. (2003). What do we know about disability beneficiaries’
work and use of work incentives prior to Ticket? In K. Rupp & S.H. Bell (Eds). Vocational
rehabilitation: Will provide incentives work for Ticket to Work. Washington, DC: Urban Institute.
O’Day, B., & Killeen, M. (2002). Does U.S. federal policy support employment and recovery for
people with psychiatric disabilities? Behavioral Sciences and the Law, 20, 559-583.
O'Brien, P.C. (1983). The appropriateness of analysis of variance and multiple-comparison
procedures. Biometrics, 39, 787-788.
Parks, J., Svendsen, D., Singer, P., & Foti, M. (Eds.) (2006). Morbidity and mortality in people with serious
mental illness. Technical report. Alexandria, VA: National Association of State Mental Health
Program Directors (NASMHPD) Medical Directors Council.
Pincus, H.A., Kennedy, C., Simmens, S.J., Goldman, H.H., Sirovatka, P., & Sharfstein, S.S. (1991).
Evaluating disabilities based on mental impairment: APA’s formative evaluation. American
Journal of Psychiatry, 148, 1037-1043.
Porter, A., Smith, J., Payette, A., Tremblay, T., & Burt, R (2009). SSDI $1-for-2 benefit offset pilot
demonstration: Vermont pilot final report. Contractor's report to the Social Security
Administration (Contract Number SS00-05-60009). Waterbury, VT: Vermont Division of
Vocational Rehabilitation.
Proschan, M.A., & Waclawiw, M.A. (2000). Practical guidelines for multiplicity adjustment in clinical
trials. Controlled Clinical Trials, 21(6), 527-539.
Rapp, C.A., & Goscha, R.J. (2004). The principles of effective case management of mental health
services. Psychiatric Rehabilitation Journal, 27, 319-333.
Rapp, C.A., Goscha, R.J., & Carlson, L.S. (2010). Evidence-based practice implementation in
Kansas. Community Mental Health Journal, 46, 461-465.
Resnick, S.G., & Bond, G.R. (2001). The Indiana Job Satisfaction Scale: Job satisfaction in
vocational rehabilitation for people with severe mental illness. Psychiatric Rehabilitation Journal,
25, 12-19.
Rinaldi, M., Killackey, E., Smith, J., Shepherd, G., Singh, S., & Craig, T. (2010). First episode
psychosis and employment: A review. International Review of Psychiatry, 22, 148-162.
Rothman, K.J. (1990). No adjustments are needed for multiple comparisons. Epidemiology, 1, 43-46.
Rush A.J., Crismon, M.L., Kashner, T.M., Toprac, M.G., Carmody, T.J., Trivedi, M.H., et al. (2003).
Texas Medication Algorithm Project, Phase 3 (TMAP-3): Rationale and study design. Journal
of Clinical Psychiatry, 64, 357-369.
References-7
Ryan, C.S., Sherman, P.S., & Bogart, L.M. (1997). Patterns of services and consumer outcome in an
intensive case management program. Journal of Consulting and Clinical Psychology, 65(3), 485-493.
Salyers, M.P., Becker, D.R., Drake, R.E., Torrey, W.C., & Wyzik, P.F. (2004). Ten-year followup of
clients in a supported employment program. Psychiatric Services, 55, 302-308.
SAMHSA. (2009). 2009 CMHS uniform reporting system
http://www.samhsa.gov/dataoutcomes/urs/urs2009.aspx.
output
tables.
Available
at:
Schimmel, J., Stapleton, D.C., Song, J. How common is "parking" among Social Security Disability Insurance
(SSDI) beneficiaries? Evidence from the 1999 Change in the Level of Substantial Gainful Activity (SGA).
Working Paper 220. University of Michigan, Michigan Retirement Research Center.
http://www.mrrc.isr.umich.edu/publications/Papers/pdf/wp220.pdf.
Schochet, P.Z. (2008). Technical methods report: Guidelines for multiple testing in impact evaluations. NCEE
2008-4018. U.S. Department Of Education.
Shafer, M.S., & Huang, H.W. (1995). The utilization of survival analysis to evaluate supported
employment services. Journal of Vocational Rehabilitation, 5, 103-113.
Social Security Administration. (2002-2009). Annual statistical report on SSDI. Retrieved 11/29/10.
http://www.socialsecurity.gov/policy/docs/statcomps/.
Social
Security Administration. (2008). Disability evaluation under Social Security.
http://www.ssa.gov/disability/professionals/bluebook/. Retrieved November 29, 2009.
Social Security Administration. (2009). 2009 RED BOOK: A summary guide to employment supports for
individuals with disabilities under the Social Security Disability Insurance and Supplemental Security
Income Programs.
Stapleton, D., Liu, S., Phelps, D., & Prenovitz, S. (2010). Work activity and use of employment supports
under the original Ticket to Work regulations: Longitudinal statistics for new Social Security Disability
Insurance beneficiaries, final report. Washington DC: Mathematica Policy Research.
Suppes, T., Rush, A.J., Dennehy, E.B., Crismon, M.L., Kashner, T.M., Toprax, M.G., Carmody, T.J.,
et. al. (2003). Texas Medication Algorithms Project, phase 3 (TMAP-3): Clinical results for
patients with a history of mania. Journal of Clinical Psychiatry, 64, 370-382.
Taylor, A.C., Bond, G.R., Tsai, J., Howard, P.B., El-Mallakh, P., Finnerty, M., et al. (2009). Scales to
evaluate quality of medication management: Development and psychometric properties.
Administration and Policy in Mental Health and Mental Health Services Research, 36, 247-254.
Thompson, J.R. (1998). Comment on “Multiple comparisons and related issues in the interpretation
of epidemiologic data” (1995V142 p904-908). American Journal of Epidemiology, 147, 801-806.
Thompson, J.R. (1998). Reply to comments on “Multiple comparisons and related issues in the
interpretation of epidemiologic data”' (1998V147 p807-814). American Journal of Epidemiology,
147, 815-815.
References-8
Torrey, W.C., Bond, G.R., McHugo, G.J., & Swain, K. (2011). Evidence-based practice implementation in
community mental health settings: The relative importance of key domains of implementation activity.
Administration and Policy in Mental Health and Mental Health Services Research, DOI:
10.1007/s10488-011-0357-9.
Trivedi, M.H., Rush, A.J., Crismon, M.L., Kashner, T.M., Toprac, M.H., Carmody, T.J., et. al. (2004).
Clinical results for patients with major depressive disorder in the Texas Medication
Algorithm Project. Archives of General Psychiatry, 61, 669-680.
Twamley, E.W., Narvaez, J.M., Becker, D.R., Bartels, S.J., & Jeste, D.V. (2008). Supported
employment for middle-aged and older people with schizophrenia. American Journal of
Psychiatric Rehabilitation, 11, 76-89.
U.S. Bureau of Labor Statistics, Department of Labor. (2011). Unpublished tables from the Current
Population Survey, 2010 annual averages.
U.S. Bureau of Labor Statistics, Department of Labor (2010). Persons with a disability: Labor force
characteristics – 2009. http://www.bls.gov/news.release/disabl.nr0.htm. Released August 25,
2010.
U.S. General Accountability Office. (2002, January). SGA levels appear to affect the work behavior of
relatively few beneficiaries, but more data needed. Report to the Committee on Finance, U.S. Senate
and the Committee on Ways and Means, House of Representatives.
Uttaro, T., & Mechanic, D. (1994). The NAMI consumer survey analysis of unmet needs. Hospital
and Community Psychiatry, 45, 372-374.
Van Dongen, C.J. (1996). Quality of life and self-esteem in working and nonworking persons with
mental illness. Community Mental Health Journal, 32, 535-548.
Velligan, D.I., Diamond, P., Wang, M., Glahn, D.C., Castillo, D., Bendle, S., et al. (2007).
Relationships among subjective and objective measures of adherence to oral antipsychotic
medications. Psychiatric Services, 58, 1187-1192.
Ware, J.E., Jr., Kosinski, M., Turner-Bowker, D.M., & Gandek, B. (2002). How to score version 2 of the
SF-12® Health Survey (with a supplement documenting version 1). Lincoln, RI: QualityMetric Inc.
Ware, J., Kosinkski, M., and Keller, S. (1996). SF-12: A 12-item short-form health survey. Medical
Care, 34(3), 220-233.
Zimmerman, D.W. An extension of the Rank Transformation Concept (1992).
Experimental Education, 61(1), 73-80.
References-9
Journal of
Glossary of Terms
Term
Definition
Administrative Drop
Treatment group participants who were removed
from the study for failing to complete the General
Medical Exam (GME), failure of the GME, or
inability to meet an initial study requirement for
participation in the treatment condition (e.g., moved
out of the country).
Affective Disorder
A psychiatric disorder characterized by a disturbance
of mood, accompanied by a full or partial manic or
depressive syndrome (i.e., bipolar or depression).
Benefits Counseling
Personalized information about the impact of work
(earnings) on Social Security benefits, other
entitlements, or income-based assistance program
benefits provided to a beneficiary by a trained
benefits counselor.
Case Management
A clinical advocacy service designed in collaboration
with the participant and treatment staff that provides
for coordination of and access to a wide range of care
needed to improve physical, social, and mental
functioning.
Catchment Area
Refers to the specific postal zip codes included in the
communities, cities, or counties that the study site
typically served; used to define the geographic area
from which SSDI beneficiaries were recruited into the
study.
Collaborative Care Model
A model of care in which a trained non-physician
clinician works in tandem with prescribers.
Competitive Employment
Jobs that pay at least a minimum wage are "owned"
by the employer (rather than a mental health center or
rehabilitation agency), and are not set aside positions
for people with disabilities.
Comprehensive Coverage
In the MHTS, refers to health insurance coverage
equivalent to Medicare Parts A, B, and D.
Continuing Disability Review (CDR)
A periodic review (by the Social Security
Administration) of complete current information
about a beneficiary’s condition to determine
continued eligibility to receive SSDI and/or SSI
benefits.
Glossary-1
Term
Definition
Co-occurring Condition (Psychiatric)
An active drug and/or alcohol diagnosis with
simultaneous presentation of a thought or affective
disorder. Often referred to as a dual diagnosis.
Creditable Coverage
Health coverage you have had in the past (such as a
group health plan [including COBRA continuation
coverage], an HMO, an individual health insurance
policy, Medicare or Medicaid), and this prior coverage
was not interrupted by a significant break in coverage.
The time period of this prior coverage must be
applied toward any pre-existing condition exclusion
imposed by a new health plan. Proof of creditable
coverage may be shown by a certificate of creditable
coverage or by other documents showing an
individual had health coverage, such as a health
insurance ID card.
Donut Hole
A Medicare Part D annual coverage gap that occurs
when a beneficiary has spent a certain amount of
money for covered drugs through his or her plan.
During this coverage gap, the beneficiary must pay all
costs out-of-pocket for his or her prescription
medications (up to a limit). The yearly deductible,
coinsurance or copayments, and payments in the
coverage gap all count toward this out-of-pocket
limit. The limit doesn’t include the drug plan’s
premium.
Dual-eligible
A beneficiary who is eligible for disability benefits
under both the Social Security Disability Insurance
(SSDI) and Supplemental Security Income (SSI)
programs.
Employment Specialist
Specially trained staff who provide employment
services (such as engagement, vocational profiling,
job search, job development, and individualized
follow-up support) to individuals with serious mental
illness wanting to return to work.
Engagement
Beneficiary level of participation in the treatment
intervention components of the MHTS. Components
measured include supported employment (SE), other
behavioral health (OBH) services, and systematic
medication management (SMM).
Essential Work-Related Expense (WRE)
An item or expense that a provider team considered
to be essential to the participant’s return to work,
needed immediately, and without which work could
not commence.
Glossary-2
Term
Definition
Evidenced-Based Practice
A practice which, based on experimental research
findings, or based on expert consensus opinion about
available evidence, is expected to produce a specific
clinical outcome in the tested population.
Fidelity (Model)
Adherence to the principles of an evidence-based
program model.
•
Program Level Fidelity
In the MHTS, the extent to which a study site
provided supported employment services consistent
with the Individual Placement and Support model.
For the MHTS, this was referred to as “Study Site
Fidelity.”
•
Individual Level Fidelity
The extent at which an individual received SE and
other behavioral health and related services that were
expected to be received. For the MHTS, this was
referred to as “Beneficiary-level Fidelity.”
Health Benefits Plan (HBP)
Refers to the benefits package provided to MHTS
treatment group participants, consisting of the
provision of healthcare (including health insurance
coverage) and supported employment services.
Individual Placement and Support Model
(IPS)
An evidence-based model of supported employment
that adheres to core principles, specifically designed
to serve individuals with serious mental illness.
Intent to Treat (ITT)
A fundamental principle in conducting randomized
trials where all randomized participants are analyzed
according to the group to which they were randomly
assigned regardless of extent of adherence to the
treatment intervention, level of engagement, or
whether they withdrew from the study.
Medicaid
Federal and state (jointly-funded) health insurance
program for low-income and needy individuals. It
covers certain children, some or all of the aged, blind,
and/or disabled individuals in states eligible to receive
federally assisted income maintenance payments. Title
XIX of the Social Security Act authorizes Medicaid.
The law gives the states options regarding eligibility
under Medicaid. Medicaid is operated under rules set
forth by the U.S. Centers for Medicare and Medicaid
Services.
Glossary-3
Term
Definition
Medicare
A federal health insurance program for people age 65
or older and people with disabilities under age 65
who meet specific requirements. SSDI beneficiaries
are eligible for Medicare after a 24-month waiting
period. Medicare is operated by the U.S. Centers for
Medicare and Medicaid Services. Title XVIII of the
Social Security Act authorizes Medicare. Medicare
consists of hospital insurance (Part A), supplemental
medical insurance (Part B), Medicare Advantage (Part
C), and voluntary prescription drug benefits (Part D).
Other Behavioral Health (OBH) and
Related Services
Refers to the range of services that were provided to
MHTS treatment group participants in addition to
SMM and SE services for the treatment,
rehabilitation, or habilitation of mental disorders.
These services include therapy, case management,
general medical care, social skills training, financial
assistance, housing assistance, substance abuse
treatment, family counseling, and legal assistance.
Psychiatric Impairment
Illness based on signs or symptoms of distress and/or
dysfunction relating to thought, emotion, or behavior.
Psychosocial Interventions
Treatment aimed at improving an individual’s
interactions with others and the environment.
Examples include Assertive Community Treatment,
Cognitive Behavioral Therapy, Dual Diagnosis
Treatment of Mental Illness and a Substance Abuse
Disorder, Family Education and Support, illness selfmanagement, and supported employment.
Quality Assurance (QA)
For the MHTS, QA is a periodic individual chart
review to assess the quality of systematic medication
management by the NCC and the prescriber. SMM
QA consists of 23 items scored from 1 to 5. Not all
items are applicable to all charts. The summary score
is the sum of all scores divided by the number of
scored items multiplied by five.
Quality Management (QM)
For the MHTS, QM is a regularly scheduled
individual chart review to a) monitor needs, b)
monitor services received, and c) identify participants
in need of improved supported employment,
behavioral health, and/or systematic medication
management services.
Glossary-4
Term
Definition
Repayee
A person, agency, organization or institution selected
by SSA to manage a beneficiary’s funds when it has
been determined that the beneficiary is unable to do
so on his/her own.
Schizophrenia
A chronic mental disorder characterized by
disintegration of thought processes and of emotional
responsiveness. It most commonly manifests as
auditory hallucinations, paranoid or bizarre delusions,
or disorganized speech or thinking and it is
accompanied by significant social or occupational
dysfunction.
Severe Mental Illness
A mental disorder associated with a significant degree
of distress and dysfunction in at least one area of life
functioning (e.g., work, school, daily care, etc.).
Social Security Disability Insurance (SSDI)
A disability insurance entitlement program
administered by the Social Security Administration
authorized under Title II of the Social Security Act.
SSDI provides benefits (income and health insurance)
to disabled or blind individuals who are “insured” by
workers’ contributions to the Social Security trust
fund. These contributions are based on earnings
(individual, spouse, or parents) as required by the
Federal Insurance Contributions Act (FICA).
Eligibility for SSDI is based on SSA’s disability
criteria, which includes inability to engage in any
substantial gainful activity (SGA) because of a
medically-determinable physical or mental
impairment.
SSA Mental Disorders
The nine categories of mental disorders by which
SSA considers an individual for disability eligibility.
The categories include the following: Organic mental
disorders (12.02); schizophrenic, paranoid, and other
psychotic disorders (12.03); affective disorders
(12.04); mental retardation (12.05); anxiety-related
disorders (12.06); somatoform disorders (12.07);
personality disorders (12.08); substance addiction
disorders (12.09); and autistic disorder and other
pervasive developmental disorders (12.10).
Substance Use Disorder
A maladaptive pattern of drug or alcohol use and/or
dependence with adverse physical and/or emotional
consequences.
Glossary-5
Term
Definition
Substantial Gainful Activity (SGA)
A term used by SSA to describe a level of work
activity and earnings. Work is “substantial” if it
involves doing significant physical or mental activities
or a combination of both. For work activity to be
substantial, it does not need to be performed on a
full-time basis. Work activity performed on a parttime basis may also be substantial gainful activity.
Work is “gainful” if a) it is performed for pay or
profit, b) it is of a nature generally performed for pay
or profit, or c) it is intended for profit, whether or
not profit is realized. At the start of the MHTS in
2006, earnings averaging more than $860 for nonblind beneficiaries demonstrated SGA. At the end of
the MHTS in 2010, earnings averaging more than
$1,000 for non-blind beneficiaries demonstrated
SGA. SGA is used as a factor to determine eligibility
for benefits.
Supplemental Security Income (SSI)
An income supplement program administered by the
Social Security Administration authorized under Title
XVI of the Social Security Act. The SSI program
makes cash assistance payments to aged, blind, and
disabled individuals (including children) who have
limited income and resources. The federal
government funds SSI from tax revenues. Many
states pay a supplemental benefit to individuals in
addition to their federal benefits.
Supported Employment (SE)
Services intended to assist people with disabilities
participate as much as possible in the competitive
labor market, working in jobs they prefer, and with
the level of professional help they need.
Systematic Medication Management (SMM)
An algorithmic method to facilitate and promote
prescriber use of evidence-based guidelines and
recommendations for medication management for
individuals with schizophrenia or an affective
disorder.
Ticket to Work (TTW)
Program administered by the Social Security
Administration (SSA) that issues tickets to eligible
beneficiaries who, in turn, may choose to assign those
tickets to an Employment Network of their choice to
obtain employment services, vocational rehabilitation
services, or other support services needed to obtain
or keep a job. It is a free and voluntary service. There
is no penalty for not using an assigned ticket.
Glossary-6
Term
Definition
Title II Beneficiary
An individual who meets the eligibility criteria set
forth under Title II of the Social Security Act to
receive SSDI benefits.
Vocational Plan
Individualized set of written procedures aimed at
assisting an MHTS treatment group participant to
reach his or her vocational goal.
Vocational Profile
A template to document a beneficiary’s work history,
job preferences, and other factors that would
influence the vocational plan.
Vocational Rehabilitation (VR)
Any service or program that has the goal of assisting
individuals with disabilities’ return to work or attain
self-sufficiency.
Westat Insurance Planner (WIP)
A Westat staff member who is responsible to oversee
the MHTS insurance services, including plan
eligibilities and access, and insurance-related
disbursements provided to or on behalf of
participants randomized to the treatment arm of the
study.
Withdrawal
Any participant who chose to formally end his or her
participation in the MHTS prior to his or her 24th
month of enrollment in the study.
Glossary-7
List of Acronyms
Acronyms used by the Mental Health Treatment Study
Acronym
ASI
BACS
BMI
BPAO
CAPI
CDR
CHAMP-VA
CMS
DHHS
DSM-IV
EBP
EIR
EROC
FICA
GED
GME
HBP
HFS
IPS
IRB
ITT
IVR
JTPA
MBR
MDPF
MEF
NCC
NHIS
OBH
PIA
PORT
QA
QM
QMPD
QOLI-M
RA
RFP
Full Term
Addiction Severity Index
Brief Assessment of Cognition in Schizophrenia
Body Mass Index
Benefits Planning, Assistance, and Outreach Program
Computer-Assisted Personal Interview (Baseline, Quarterly, Final Followup)
Continuing Disability Review
Civilian Health and Medical Program of the Department of Veterans Affairs
Centers for Medicare and Medicaid Services
Department of Health and Human Services
Diagnostic and Statistical Manual of Mental Disorders, 4th Edition
Evidenced-Based Practice
Employment and Income Review
Electronic Record of Contact
Federal Insurance Contributions Act tax
General Equivalency Diploma
General Medical Exam
Health Benefits Plan
Hirsch Financial Services
Individual Placement and Support
Institutional Review Board
Intent to Treat
Interactive Voice Response (Westrax)
Job Training Partnership Act
Master Beneficiary Record
Medicare Drug Plan Finder
Monthly Encounter Form
Nurse Care Coordinator
National Health Interview Survey
Other Behavioral Health
Primary Insurance Amount
(Schizophrenia) Patient Outcomes Research Team
Quality Assurance
Quality Management
Quality Management Project Directors
Lehman Quality of Life Inventory - Modified
Research Assistant
Request for Proposal
Acronyms-1
List of Acronyms (continued)
Acronym
RIG Meeting
RSI
SAMHSA
SCAP
SCID
SCHIP
SE
SE Specialist
SGA
SMM
SMS
SSA
SSDI
SSI
SURF
TANF
TMAP
TPQY
TRC
TTW
UCDI
UMBC
UTHSC
VA
VO
VR
WIP
WIPA
WRE
Full Term
Recruitment Information Group Meeting
Retirement and Survivors Insurance
Substance Abuse and Mental Health Services Administration
The Schizophrenia Care and Assessment Program Health Questionnaire
Structured Clinical Interview for DSM Disorders (DSM-IV)
State Children’s Health Insurance Program
Supported Employment
Supported Employment Specialist/Employment Specialist
Substantial Gainful Activity
Systematic Medication Management
Study Management System
Social Security Administration
Social Security Disability Insurance
Supplemental Security Income
Services Utilization and Resources Form
Temporary Assistance for Needy Families
Texas Medication Algorithm Project
Third Party Queries
Telephone Research Center (Westat)
Ticket to Work
Uniform Client Data Inventory
University of Maryland Baltimore County
University of Texas Health Science Center at San Antonio
Veterans Affairs
ValueOptions
Vocational Rehabilitation
Westat Insurance Planner
Work Incentives Planning and Assistance
(Essential) Work-Related Expense
Acronyms-2
Acknowledgement of Study Site Staff
The MHTS investigators would like to thank the Supported Employment staff at each of our 23
MHTS study sites. In addition, we would like to recognize and thank the following administrators,
staff, Nurse Care Coordinators (NCCs), Research Assistants (RAs), and other researchers at each
site:
Aiken Barnwell Mental Health Center
Aiken, SC
John Young*
Rick Acton
Tamara Smith
Julia Ellis (NCC)
Shannon Britt (RA)
Adult & Child Mental Health Center, Inc.
Indianapolis, IN
Michael McKasson
Susan Reul (NCC)
Kimberly Dreher (RA)*
Abigail Mook (RA)
Donna Hofmann (NCC Assistant)*
Boley Centers
St. Petersburg, FL
Gary MacMath
Miriam Williams
Curtis Anderson
Junior Bentham (NCC)
Christine Pickens (NCC)
Michael Haines (RA)*
The Center for Health Care Services
San Antonio, TX
Mary Bohne*
James Castro
Linda Lopez
Sheryl Pavlick (NCC)*
Nancy Howland (NCC)
Michael Gambuti (RA)*
Steven Parra (RA)
Freda Facey (NCC Assistant)
Cobb/Douglass Community Services
Board
Smyrna, GA
Tod Citron
Doug Crandell
Michelle Robison
Christine Wright (NCC)*
Kelly Smith (NCC)
Jennie Keleher (RA)
Camille McDonald (NCC Assistant)
Columbia River Mental Health Services
Vancouver, WA
Gregory Robinson*
Nancy Parker
Lisa Shannon
Caroline Wrightman (NCC)*
Laryn Dole (RA)*
Sarah Friel (RA)*
Community Connections
Washington, DC
Maxine Harris
Helen Bergman
Kate Boucek
Cheryl Reidy (NCC)*
Susan Cummins (RA)*
Fayette Companies Human Service
Center
Peoria, IL
Michael Boyle
Becky Weaver (NCC)
Julie Sprouse (RA)
*Former site director or former MHTS study site research staff.
Acknowledgement-1
Henderson Mental Health Center
Ft. Lauderdale, FL
Pamela Galan*
Steven Ronick
Mayra Reimer
Linda Ciani (NCC)
Elise Ward (NCC)
Dacia Stone (RA)*
Kara Williams (RA)
Options for Southern Oregon
Grants Pass, OR
Jeffrey Krolick
Judith Klein (NCC)
Crystal McMahon (RA)*
Dan Harper (RA)*
Pathways to Housing, Inc.
New York, NY
Sheryl Silver*
Sam Tsemberis
Georgia Boothe
Joanna Price (NCC)*
Ollie Pianteri (NCC)
Ed Nolan (NCC)
Allison Zalinger (RA)
LifeWorks, NW
Portland, OR
Phyllis Maynard
Maqsuda Kabir (NCC)
Trinity Wilson (NCC)
Anthony Kramer (RA)*
Mari Meyer (RA)
Mental Health Center of Denver
Denver, CO
Roy Starks
Mary Khoury (NCC)
Shawna McGuckin (RA)*
The Mental Health Center of Greater
Manchester
Manchester, NH
Margaret Almeida*
Peter Janelle
Harry Cunningham
Amy Szczechowicz
Susan Guarino (NCC)
Mary Ellen Smith (RA)*
John Moscarito (RA)
Selma Deyak (RA)
Neighboring
Mentor, OH
Kenneth G. Gill
Jennifer Perkins (NCC)
Karen Knavel (RA)*
Robin Deemer (RA)
RISE, Inc.
Spring Lake Park, MN
Don Lavin
Joan Distler
Sheri Walczak (NCC)
Jennifer Gajewski (RA)
Jeannie Rhinhart (NCC Assistant)
South Middlesex Opportunity Council,
Inc.
Framingham, MA
Jeffrey Handler
Laurie Abruzzi (NCC)
Meagan Ray-Windrush (RA)*
Southeastern Mental Health Authority
Norwich, CT
William Newkirk
Stephenie Guess
Claire Burchfield*
Kathryn Hatfield (NCC)
Faraz Wasti (RA)*
Kevin Larson (NCC Assistant)*
*Former site director or former MHTS study site research staff.
Acknowledgement-2
Southwest Connecticut Mental Health
Systems
Bridgeport, CT
Rebecca Allen*
James Pisciotta
Celeste Cremin-Endes
Daniel Wartenberg
Kathleen Hammond (NCC)
Howard Goldberg (RA)*
Avis Harper-Brooks (RA)*
St. Luke’s House
Bethesda, MD
Cindy Ostrowski*
Larry Abramson
Karen Mann (NCC)*
Kasaan Holmes (RA)*
Lauren Matthews (RA)*
Mandi Browning (NCC Assistant)
Joan Clement (NCC Assistant)
The Thresholds (North and South)
Chicago, IL
Virginia Fraser
Cynthia Kelly (NCC)*
Ann Naumes (NCC)
Laurel Lipkin (NCC)
Susan Kaiser (RA)*
Chad Cancro (RA)
Isabela Krzos (RA)
Wyandot Center for Community
Behavioral Health, Inc.
Kansas City, KS
Leslie Young
Dena White (NCC)*
Mica Willis (NCC)
Christine Swenson (RA)
*Former site director or former MHTS study site research staff.
Acknowledgement-3